Files
craft-anything/docs/ROADMAP.md
BFlat e4ba479004 Merge pull request #1991 from Plastic-time/bugfix/r3.13-html-incomplete-tags
A capture cut off inside a tag, such as one ending in `<script src=app.js`, reported a script no browser would run; cut-off link, title and meta tags could supply facts too. Page evidence now needs the tag to be closed.
2026-11-17 10:28:06 +07:00

268 KiB
Raw Permalink Blame History

Roadmap

The shared plan for reaching the product described in the README, for people and agents alike. docs/STATUS.md says what works today; this file says what comes next, who owns it and what an agent may pick up; roadmap-history.md keeps what's finished. Rebuilt on 2026-11-02 from main at 2e4afc5, with v0.3.0 the latest release.

How to use this file

  • Each item has an id (R<milestone>.<n>), an owner, a status and a list of pending changes. Pending changes are small, independently mergeable units. Ids are never reused or renumbered: commits, code comments and PRs quote them.
  • Claiming: before you start, put — claimed: <name>, branch <branch> on the pending change and push your branch early (or open a draft PR) so others see it. Don't take a change someone else has claimed. — in PR #n marks one a PR already covers: review that PR rather than starting again. The full working rules are in AGENTS.md ("Working rules", mandatory). Items owned by maintainers are steered by the project maintainers: agents may take their unclaimed pending changes, but design decisions marked decision wait for a maintainer.
  • Finishing: tick the change ([x]) where it is, add the commit or PR, update docs/STATUS.md if what works changed, and add a line to log/devlog.md. Maintainers move ticked lines to roadmap-history.md when a release ships.
  • New work: add a - [ ] line under the item it belongs to; a new item takes the next free id.
  • Status: todo · in progress · blocked (<why>) · done.
  • Follow AGENTS.md (layering, never crash, everything is a tool, platform::process for every external program) and the architecture rules below.

Architecture rules for new work

The project should be easy to fork and bend. New features land behind a seam, not inside a match:

  1. Traits plus a registry, not enums. Target kinds, checks, builders, asset providers and external engines are traits with a registry that holds built-in implementations. Adding one means adding a file and one registration line (as tools already work).
  2. Data over code for catalogues. Stacks, game-engine build recipes, model ids and prices, and engine install pins live in typed tables (or TOML files embedded with include_str!), not in string matches or prompt text.
  3. Headless first; flows are tools. Everything works without the GUI. Anything a person can start from the UI (rebuild, follow-ups, answering questions, engine installs, launching a result) is reachable through Engine::call and the CLI, so MCP, agents, scripts and other frontends get it for free. Orchestration, persistence and process control never live in ui-egui or apps/*.
  4. Versioned files. Every file we write and read back (craft.json, config.toml, project.toml) carries a version, has a JSON schema, and has a migration path.
  5. Structured results. Decisions such as "did the build pass" read typed fields, never English text prefixes.

Milestones at a glance

Milestone Goal Owner Where it stands (2026-11-02)
M0 Foundations Docs match the code, licences checked, one live end-to-end run maintainers Done
M1 Extensibility Seams for target kinds, checks, builders, asset providers, engines, plugins maintainers Done, except engine pins (R1.5)
M2 Product completeness Every kind of target gets study → build → check → refine with a real check maintainers Websites complete; apps and games build and run; API-key builds have never run live (R5.5); agents other than Claude Code are only partly wired (R2.2)
M3 Engine depth More tools, and the known partial ones open 314 tools; the open lane
M4 Release Packaging, signing, a managed tier maintainers v0.3.0 released, signed, on every platform; a macOS CLI question and the managed tier remain
M5 Works for real people What people hit now that the project is public maintainers Comes first

Open work

Every unticked - [ ] line in this file is open work, under the item it belongs to. cargo xtask pending lists them all, grouped by item (--count for the totals). When you find work that's missing, add it as a - [ ] line under the right item in the same PR, rather than leaving it in a PR description or a chat.

Order and dependencies

Work in phases. Items in the same phase can run in parallel. A phase says what comes first; what stops an item from starting is its Depends on column, which must be done first. A soft dependency should come first but doesn't block. Issues are ranked by people affected and how badly, after leaving out what's already fixed.

Phase Item Depends on Why here
0 Ship R5.1 The next release — The Windows start-up crash and a build-breaking resume bug are fixed on main but not released
0 R5.2 Keeping contributions moving — main isn't protected, first-time contributors' CI is held, and a waiting PR fixes a safety bug
1 Builds that work R5.3 A builder that can't make progress stops — Costs people money or quota on every backend; small
1 R5.4 Local models can build R5.3 (soft) Largest open gap, with no fix anywhere yet
1 R2.2 Every builder gets the same tools (that group), and Approvals decision for Approvals Agents other than Claude Code start but can't do the job, and Clean room's rule doesn't reach them
1 R5.5 API-key builds fail in words people can use — This path has never been run live, and its errors say nothing
1 R5.6 Tools report only what they saw — Confirmed false successes on macOS; golden rule 3
2 Proof R5.7 Live checks owed — Built but never run for real; each can reopen work above. Lines run when the machine or key they need is at hand, Clean room and the quick start first
3 Depth The rest of R2.2 — Detection waits for reports from 0.3.0 or later; ACP after the adapters work
3 R2.4 Runtime study, R2.5 Behaviour checks scripted input (R2.4) before R2.5's sheet check Makes "ready" mean more for apps and games
3 R2.7, R2.8 — Small gaps left in finished work
3 R2.9 Everyday use, R2.10 Pinned builds — Open to anyone at any time; most lines stand alone
3 R1.5 — Small leftovers
Any M3: R3.3, R3.9, R3.11 — Open lane
Later R4.1 macOS CLI, R4.2 Managed tier decision Not before M5

M5 Works for real people

R5.1 The next release

Owner: maintainers · Status: todo Two reported failures, the Windows start-up crash (#414) and a builder resuming another's session (#471), are fixed on main and still broken for everyone on 0.3.0. This is the cheapest help we can give anyone, so nothing below should hold it up for long.

  • #490 merged (549ae5e) with the CLI fix the review asked for: a string-or-object input is any, so a JSON object given to --left stays an object (tested both ways). What it reports is 0.2.0's, fixed by #93 (f6725c2): on Windows the published 0.3.0 CLI with Claude Code 2.1.179 mounted 314 of 315 tools (the 315th is the approval tool, which Claude Code doesn't list but calls), and a real Deep build reached Ready with core_action_approve answering five prompts. The reporter was asked for --version. macOS is still owed (R5.7)
  • Cut the next release (docs/releasing.md). The version is 0.4.0 (#526) and main at fb7ac86 is green, but pushing release is refused: its ruleset blocks every update and lists no one who may bypass it. A maintainer who can push it (or adds a bypass) goes next. 0.3.0 lacks 0748ef1 (#414: Windows starts DX12 only, the crash behind "it won't open"), 1e8f661 (#471: a builder never continues another builder's session) and 3357daf (#430: Claude Code's message on stdin)
  • See that the launcher offers the new version: #74, #104 and #420 installed through it, and nothing in this repository says how it picks one
  • After it ships, close what it fixes: #166, #346, #420, #478. #74, #104 and #222 give no detail, so they may have another cause: keep them open until their fixes are validated against the published release, following docs/releasing.md. Ask #201 whether 3861132 (#379) ended its early compression (a guess at the link)
  • Close the tickets 0.3.0 already fixed, and the spam: #47, #53, #66, #95, #112, #113, #130, #133, #138, #152, #179, #182, #185, #202, #214, #329, #344, #377, #378, #400, #449; spam #194, #195. #113, #329 and #377 are 0.2.0's missing CLI, fixed by 1b54cbf, not what #490 reports. Not yet: #64 (its cause is fixed, its retries aren't: R5.3), #142 (R5.3), and #462, which says none, not #490's failure, and gives no version: ask for it
  • Sometimes starting the app shows nothing: after the release, see from desktop.log whether it still happens. DX12-only may have ended it; a GOG Galaxy Vulkan overlay layer was seen loading into the app
  • A program that is found but can't start (an npm script whose node is missing) says where it is and what to look at, not "isn't installed" (#152, #95); on Windows a program found is started by its path, not its name
  • Live-check #152 on Windows 11 with OpenCode from npm, and on a machine with the agent but no Node.js
  • A start that dies before its window (a native driver crash leaves no dialog or log, #74, #104, #33) is remembered: starting beside desktop.log is written before the window is asked for and removed once it's open, and a start that finds it draws with WARP that once (apps/craft-anything/src/start_guard.rs)
  • Live-check the start fallback on Windows: make a start die before its window (kill the process at launch), start again, and see desktop.log say it drew without the graphics card and log drawing with an adapter of type Cpu (WARP), and that the window opens. Passed on 2026-11-04 (main at 869eb85, release build, Windows 11, RX 6900 XT): a start killed right after drawing with the Radeon left starting; the next logged "the last start never opened its window", drew with "Microsoft Basic Render Driver" (Cpu, DX12), logged window open and removed starting — claimed: 00xJS, branch docs/r5.1-start-fallback-live-check
  • The software retry still lets wgpu start every DX12 adapter before pick_adapter runs, so a driver that crashes while its device is created crashes the retry too. wgpu has no way to skip hardware adapters; needs a different route (another renderer, or a helper that starts the app with only WARP visible)
  • #794: on some systems DX12 refuses egui's own pipeline (egui_pipeline, 0x80070057 from CreatePipelineState) on AMD, Intel and NVIDIA cards and on WARP too (lightcraft#812, pdfcraft#1046, wgpu#10771), so the software retry failed the same way every start, and the dialog only said to update the driver. Each failed start now moves on to the next renderer (DX12 → WARP → Vulkan → GL, apps/craft-anything/src/renderer.rs), keeps the one that worked for this app version, and starts again by itself after a wgpu error. Checked with a build that failed every DX12 start; the real failure wasn't reproduced (STATUS.md) — claimed: 00xJS, branch bugfix/r5.1-renderer-fallback
  • Live-check #794's fallback on a machine where DX12 refuses egui_pipeline (the reporter's 780M, or lightcraft#812's)
  • A window that fails after it was shown no longer says "The next start will draw with …": its start mark is already gone, so the next start draws as before (#909 review); the dialog says it only when the window was never shown (next_start_note) — claimed: 00xJS, branch bugfix/r5.1-retry-text-after-shown
  • Add to docs/releasing.md: before publishing, a Deep build through Claude Code from the finished package, on a Mac and on Windows. Starting the CLI from the package (#91) doesn't show whether Claude Code accepts its tools — PR #982; claimed: doubektomas, branch docs/r5.1-packaged-deep-build-check. The checklist is documented; live package runs remain release validation work.
  • Add to docs/releasing.md: after a release, close every ticket marked as waiting for it, so they don't stay open marked fixed — PR #1006; claimed: doubektomas, branch docs/r5.1-post-release-tickets. The checklist is documented; closing tickets waits for publication of the release containing their fixes.

R5.2 Keeping contributions moving

Owner: maintainers · Status: in progress

  • Fix PR #1030's Windows descendant fixture startup and stop the watched launch before startup assertions can fail; cover the failure cleanup path — claimed: Codex, branch bugfix/r5.2-watched-launch-tests
  • Run PR #1030's shell-free parent/descendant fixture and startup-timeout cleanup test on Windows; the contributor reproduced startup failure and leaked processes on 07671fe. Verified on Windows 11: cargo test -p craft-anything-platform --lib process::launch_tests passed five runs (11 passed, 6 ignored, the six being the stand-ins), stopping_a_watched_program_stops_its_descendant and a_descendant_startup_timeout_stops_the_watched_launch among them; after the runs and a 35 s wait tasklist showed no PING.EXE, and the only cmd.exe and powershell.exe were ones already running before the first run — claimed: muraa-p, branch test/r5.2-watched-launch-windows-check
  • Clarify PR #1030's test Rustdoc: post-stop heartbeat sampling, child-fixture guard, and the direct-run no-write check — claimed: Codex, branch bugfix/r5.2-watched-launch-tests
  • Address PR #1030 review: take the heartbeat baseline after stopping, guard the descendant stand-in against direct ignored-test runs, and document the test helpers — claimed: Codex, branch bugfix/r5.2-watched-launch-tests
  • regression_preview_limits_connections_and_releases_listener (platform) fails on macOS with Invalid argument in its socket fixture, also reported before PR #1030's review fixes. The fixture set its excess clients' read timeout after dropping the server, and macOS refuses socket options on a connection it has reset; the fixture now sets them when connecting — PR #1207; claimed: rmsff, branch test/r5.2-macos-preview-socket-fixture Review comes before new work while the queue is long, smallest and clean PRs first. A PR is merged only when it's green and a maintainer has fully reviewed it, code as well as wording.
  • Maintainers: protect main (PRs required, squash only); the only ruleset covers release
  • cargo xtask secrets treats every environment variable whose name contains AUTH as a secret, git's GIT_AUTHOR_NAME among them, so the pre-commit hook refuses any added line with the committer's own name when it's 12 characters or more (a roadmap claim by a contributor whose git name is their handle). Leave git's identity variables out of env_values (NOT_SECRET); OAUTH, AUTHORIZATION and other AUTH names are still checked — claimed: youkaicountry, branch bugfix/r5.2-secrets-git-identity
  • Auto mode's delete check is a safety bug that shows up as Mac test failures. It splits a command on spaces, so rm -rf "C:/x/Game Backup/y" from a project at C:/x/Game goes ahead unasked, and it lower-cases paths, so on a Mac or Linux a delete inside a project with capitals in its path is refused; auto_mode_keeps_going_but_never_does_the_dangerous_things fails on a Mac for that. real_binary_end_to_end_if_present fails there too, expecting /bin/ls to be Linux: both fixed (#357)
  • decision First-time contributors' CI: GitHub holds their runs for approval, so their PRs show no gate run (#490 and #413 were among them). Approve each push, or change the repository's approval setting
  • #193 changes the language list that #486 (merged) changed too: it needs a rebase. #468 looks redundant with #500: ask its author
  • decision Wide PRs with no roadmap line: #416 (project goals, 34 files, cites an R2.11 that doesn't exist; that id stays free for it) and #503 (two asset providers and spending controls, 37 files, cites the finished R1.4). Decide whether we want them before more is built on them
  • decision Replies don't reach people who file through the ticket relay (each such ticket says so), yet the bot asks them for details. About half the open issues came that way. Get replies back to them, or stop asking
  • which::tests::usual_windows_dirs_come_after_path (platform) failed on Windows (10 found, 9 expected, "nothing twice"): it didn't count Antigravity's second folder, fixed by #770
  • capture::console_tests::reads_what_the_program_drew_and_stops_at_the_deadline failed once with the whole platform suite running and passes alone (passed 3 whole-suite runs on Windows 11 on 2026-11-04). Find what it depends on
  • process::launch_tests::a_launcher_that_hands_off_keeps_its_group_until_stopped failed on a busy Windows machine: it killed the run 500 ms after the start, and a cmd that hadn't exited by then made a cancelled run, not a hand-off. The test now waits for the launcher's mark, and a kill that finds the program already exited reports its exit, not a cancel — claimed: 00xJS, branch test/r5.2-hand-off-race
  • The private browser exit regression checks the exit error code and the before-readiness diagnostic. Its separate 20 s stopwatch timed the entire tool call, including process startup and cleanup, and could fail independently of that diagnostic under load. The original Windows failure was not reproduced. — claimed: rudycelekli, branch test/r5.2-browser-exit-diagnostic
  • a_remote_query_expands_only_the_remote_working_directory compares the shell's $PWD with the fixture's canonical path, so macOS's /var → /private/var alias does not fail the test. Remote expansion and special-character quoting checks are retained — PR #949 — claimed: Codex, branch test/r5.2-macos-remote-query-path
  • which::tests::usual_windows_dirs_come_after_path (platform) failed on Windows with 10 found, 9 expected ("nothing twice"): #748's second Antigravity folder, counted since #770
  • capture::console_tests (platform) timed out now and then with the whole platform suite running. On Windows Server 2022 a program another thread starts while a console opens inherits its handles and holds it open until it ends, and closing a console that programs are still attached to can hang its host. Program starts and console opens (and Unix terminal opens, which have a shorter gap of the same kind) now take one lock (process::STARTING), and a timeout ends the program's whole tree before closing — claimed: SoldierPePe, branch test/r5.2-console-capture-flake
  • A console program that exits by itself and leaves programs attached to its console: closing it can still hang the console host on Windows Server 2022 (only a timeout ends the tree)
  • process::launch_tests::a_watched_program_can_be_stopped leaves its sleep 30 (Windows: ping -n 30) running, on main too: stop ends the shell but not what the shell started (#1027). Verified on Windows 11: after the test (5 runs) and after the whole craft-anything-platform suite plus a 35 s wait, tasklist showed no PING.EXE. Not checked: a panic before stop, dropping the handle without stop, Windows 10 or Server — claimed: yashpatadiya1724-cmyk, branch docs/r5.2-watched-stop-windows-check
  • a_helper_is_judged_by_whether_it_runs (ai crate) sometimes failed when several crates' tests ran at once: a script just written can't start while another thread's child still holds it open ("Text file busy"). #787 tries again: on Linux 4 of 72 runs failed before it, none after
  • Revisit the two advisory exceptions in deny.toml: rsa (no fix; we only verify with public keys) when rsa 0.10 lands, and unmaintained zbus_names/zvariant (Linux-only via keyring and accesskit) when those move on. deny.toml's comments pointed at R0.3, which never held them; they name this line now — claimed: kuseidos, branch docs/r5.2-deny-points-at-its-own-roadmap-line
  • Check whether the release workflows still run actions that target Node.js 20. None is named here any more: every job calls storytold/craft-common-actions@main, so a fix belongs there
  • Correct STATUS.md's obsolete GUI-only limit: document the shared project feed, engine setup, launch and preview tools, with their consent and lifetime limits — PR #1014 — claimed: Codex, branch docs/r5.2-headless-status
  • A spinner on screen no longer makes the window redraw at the display's rate: egui's own asks for a frame at once every frame, ours at the animation's pace (widgets::Spinner, #579)
  • A job whose progress isn't known no longer makes the window redraw at the display's rate: the jobs list used egui's animated ProgressBar, which asks for a frame at once every frame like egui's spinner did (widgets::ProgressBar) — claimed: Claude, branch bugfix/r5.2-paced-progress-bar
  • Measure #579 on Windows 11 (0.3.0 portable): the app's CPU while idle, while a build runs and during a website capture, before and after; and why a website capture is slow and sits for a long time (no details were given, so nothing was changed for it)
  • remote_discovery_uses_the_remote_program_and_connection_context fails on a computer with OpenCode installed: its made-up path is missing, so program_of falls back to opencode — claimed: justinjohn0306, branch test/remote-discovery-without-opencode
  • Idle CPU on an AMD Radeon, measured (2026-11-04, main at 869eb85, release build, Windows 11 build 26300, RX 6900 XT, driver 32.0.21045.16010, Ryzen 5 2600X; the window open on a fresh profile, nothing changing). DX12 on the card: one core at ~100% for as long as the window is open; Vulkan: the same; GL: 0%; WARP: ~280% for the first 5 s after start, then 0%. The busy thread isn't the app's: on DX12 it starts in amdxc64.dll, on Vulkan in amdvlk64.dll, and it starts with the device, not the main thread. A bare eframe window does the same with every present mode and frame latency tried (AutoVsync, Fifo, Mailbox, Immediate; latency 1, 2, 3), so nothing in our frame pacing causes or avoids it; GL alone stays idle. Seen in the same way with PdfCraft and PhotoCraft (pdfcraft#3255) — claimed: 00xJS, branch docs/r5.2-idle-cpu-amd
  • decision On the one Radeon measured (RX 6900 XT, driver 32.0.21045.16010, Windows 11 build 26300), AMD's idle driver thread costs a core on DX12 and Vulkan; other Radeons (integrated ones too), drivers and Windows builds haven't been checked. Say so in the troubleshooting docs (with WGPU_BACKEND=gl as the way out, which loads AMD's OpenGL driver), report it to AMD, or leave it
  • Limit the decision line above to what was measured (one RX 6900 XT, one driver, one Windows build) and say other Radeons and drivers aren't checked (#912 review) — claimed: 00xJS, branch docs/r5.2-radeon-idle-scope
  • STATUS.md has rows twice, kept by squash merges of PRs that edited the same row: OpenCode models and thinking (three copies that disagree about SSH), Claude desktop app without Claude Code, Websites as targets, Desktop app and CLI; docs/opencode.md has a paragraph twice. Keep the newest complete copy of each — claimed: youkaicountry, branch docs/r5.2-doubled-status-rows — PR #1226
  • Have cargo xtask refuse a STATUS.md table with two rows of the same name, so a merge that keeps both copies of an edited row fails CI instead of reaching main — claimed: Afloat16, branch test/r5.2-detect-duplicate-status-rows
  • docs/opencode.md still has one passage twice, missed by #1226: in "Configuration and execution", the older copy of the standalone-server sentence, cut off after "both" — claimed: youkaicountry, branch docs/r5.2-opencode-doubled-passage
  • Let a UI change land with its new strings in English while native speakers review their translations: cargo xtask i18n-coverage accepts strings listed as awaiting translation, and the translation PR empties the list — claimed: tanish2k09, branch chore/r5.2-awaiting-translation
  • tools::android_tools::jeb::tests::not_configured_refused_and_remote_urls fails now and then: it binds port 0, drops the listener and expects nothing to answer there, while the stand-in JEB servers of the tests beside it bind port 0 too and can be handed that port. Seen once in six whole-suite runs on Windows 11: engines_jeb_check got a stand-in's answer. Give the negative probe an address no other test can take — claimed: FuryNocturn, branch test/r5.2-jeb-refused-port
  • CI's Rust moved to 1.100, whose Cargo runs a test binary from target/<profile>/build/<package>/<hash>/out instead of target/<profile>/deps, so self_cli_path no longer found the CLI and the two rebuild tests that start it failed on main and every pull request. It now finds the CLI from that folder too — claimed: bflatastic, branch bugfix/r5.2-cli-in-new-build-dir

R5.3 A builder that can't make progress stops

Owner: open · Status: todo Every round spends the person's money or quota. One that can't do anything shouldn't start.

  • Preserve the interrupted request when an agent cannot restore its session: ACP starts a new session, or a registered CLI adapter falls back to Plain (#552 review). — claimed: Afloat16, branch bugfix/r5.3-interrupted-request-recovery-20261103

  • A build stops and says why after two builder turns in a row that changed nothing in app/, on every backend (#44, #110, #417, #481). The count is saved in the project, so it holds across runs: a run ends at 7 checks (max_rounds) but check numbers carry on, so #44's check 28 and #417's 35+ look like several runs (a guess: Try again is the only restart found, and Auto mode doesn't start runs) — implementation: 2fc2970948315b64d47359543b458509da42ab3f

  • A builder whose tools didn't start isn't resumed. turn (rebuild/mod.rs) tries 3 more times ("Picking up where it left off", #377) because worth_retrying doesn't know this error. Give it a typed signal, as Conversation::session_lost does, not one more word in that function's list. Say what failed and stop (pitfalls: never retry what can't succeed) — commit f80a16a66834

  • worth_retrying decides on English words in a message (AGENTS.md: structured results). Move its other cases (sign-in, credit, a missing install) to the same typed signal — commit f80a16a66834

  • Find what repeated #64's attempts and stop it: it logged 41, about 38 s apart, on 0.2.0, but turn allows 3 retries 30, 60 and 90 s apart, there too, so something else starts them

  • A rate limit is waited for only on agents with a session: turn retries only when agent.session_id() is set, which API backends and OpenCode's plain adapter never set (read in the code, not run). So #148 didn't reach #142, which was OpenCode. Wait and send the turn again for them too, then close #142 — commit f80a16a66834

  • A Claude Code plan's 5-hour or weekly limit that's used up stops the build at once and says when it resets; it was retried as a per-minute rate limit, failing again each time (#1919). Read from the stream's rate_limit_event (seen live with status: allowed), not the message; not seen live with a used-up plan. Other agents' plan limits are still retried

  • The turn's closing result error, which the run keeps, still says which plan limit is used up and when it resets; it fell back to the generic advice (#1931 review, after its merge) — claimed: justinjohn0306, branch bugfix/r5.3-plan-limit-closing-error — claimed: justinjohn0306, branch bugfix/r5.3-plan-limit-stops

  • A builder step on an API backend stops after 25 failed build commands in a row (none succeeding between), says so and keeps the project (MAX_FAILED_BUILDS in agent.rs, #399: 9.4M tokens in 19 minutes of the first stage, never compiled). Try again carries on

  • decision Whether a build gets a token or cost limit of its own, beyond rounds and that stop (#399): a model that changes code without ever running a build isn't caught by it

  • Check live that the stop fires on a real API-backend build and the message reads well. It fired on a local Ollama build (the same built-in loop) after two turns that changed nothing in app/ ("The build stopped after two builder turns changed no files in app/…"); an API-key build is still owed

  • #736: the build check and the stop message say when app/ is empty, so a builder that claims a build can't hide that nothing was written. v0.3.0 (the reporter's release) had no stop after two unchanged turns (#557, merged after it); main stops, and this adds the empty-folder wording to the build check (it failed first, before #749's website check) and to the stop message — claimed: claude, branch bugfix/r5.3-empty-app-loop

  • A build whose checks keep finding no working build stops after 10 in a row, across "Try again" presses, until a new request starts the count again (NOT_WORKING_CHECK_LIMIT, kept in craft.json as not_working_checks). "Try again" used to start six more rounds every time with nothing counted across runs, so #736's build went on to check 40 and 49 times. A run at the limit stops before a builder turn. The unchanged-turn guard already stops a builder that writes nothing; #736's exact Windows run (gemini-cli) is not reproduced here. An empty app/ is named as having no files yet (#771's wording) and the builder is asked to write the project and show its build's output, not to fix a build (NO_PROJECT_PROMPT). — claimed: fabox-os, branch bugfix/736-stop-empty-build-loop

  • Live-check #736 on Windows with gemini-cli: an empty app/ and a builder that claims npm run build stops after two turns (or 10 checks) with the "no files yet" note, and Try again doesn't restart it

  • A finished builder turn whose own CLI process lingers after its stream already said it was done (e.g. a stuck shutdown hook) no longer leaves the build on "Building"/"Pausing" for as long as the idle timeout allows (hours, for a builder turn): run_cli ends the turn RESULT_GRACE after its terminal result line instead of waiting for the process itself to exit. A wall or idle deadline that fires during that grace period (the result was already parsed) counts as completion, not a failure; a deadline with no parsed result still stops the turn as before — claimed: mytrade2909-design, branch bugfix/builder-stuck-after-turn-completes, PR #1426

  • A program that's stopped (time limit, pause, or leftovers after it exits) is waited for only 5 s: stopping it used a wait with no limit, the one step of a run that neither its time limit nor Pause could reach (#1438, #981: a check that sits for hours with Pause dead). Found by reading the code; the reporters' hangs are not reproduced here — claimed: justinjohn0306, branch bugfix/r5.3-stopped-program-wait

  • Confirm on a reporter's Windows machine that #1438 / #981 no longer sit at a check (their build-activity.log after the stall, and whether "still there 5 s after it was stopped" is in the app's log)

  • Launched::stop (Play's Stop) used the library's kill(), which waits with no limit, and then waited again, holding the lock its watcher and is_running take. It now sends the stop under the lock and waits outside it, 5 s at most, as other runs do (the console capture's kill() is portable-pty's, which doesn't wait). Not reproduced: no program here outlives its stop — claimed: justinjohn0306, branch bugfix/r5.3-play-stop-bounded

  • An OpenCode builder wrote its own progress notes ({"step":"done",…}) to inbox/build-progress.json, so every run after it stopped at "Couldn't read inbox/build-progress.json: unknown field step", with no way forward given. Only a build started in the person's own folder is told to leave Craft Anything's records alone: tell every builder, and say in the error that something else wrote the file and how to go on — claimed: youkaicountry, branch bugfix/r5.3-builder-wrote-progress-record

R5.4 Local models can build

Owner: open · Status: todo The built-in agent loop assumes a capable tool-calling model. Reports #24, #44, #110, #403, #417 and #498 say otherwise (#44's backend is a guess; #116 may be the same), and no local model build has been run live.

  • A live local build through Ollama, small target, to Ready or to a clear stop, and note in STATUS.md which models it worked with: Ollama 0.40.7 with qwen2.5-coder:14b on Windows (RTX 5090, 32k context, no prompt cut) stops clearly at "No project to build found in app/ yet", twice. The lines below say why
  • A builder on a local model is offered 13 tools (about 10,000 characters of descriptions) instead of 33-41 (23,000-30,000): the Build tools and the Core tools the brief names, not the study set a model explores with instead of writing (#417, LOCAL_BUILDER_CORE). Measured as text, not on a model; Ollama's default context (4k) was too small for the old list alone
  • Run #417's case live (Ollama, qwen2.5-coder:14b, "a scientific calculator") before and after this: does it call build_file_write now? If it still doesn't, the model can't build. Yes, with the three fixes below: over about 50 calls it wrote app/Cargo.toml (eframe), fixed the two errors the checks reported, and check 3 said "Build succeeded". But src/main.rs was still cargo's "Hello, world!": no calculator was written, and the screenshot check rightly failed it at the round limit (#753)
  • The system prompt told every model to call core_brief_write and core_plan_update first and core_handoff_write last, tools a lean local builder isn't offered, so the model asked only for core_brief_write and did nothing else for three checks. A lean builder's prompt leaves the memory steps out; live, the same model then used the Build tools
  • A call written after a sentence ("Let's list the files.\njson {…}") isn't run: #554 reads a reply that is only the JSON. Half the model's calls came this way and the turn ended with nothing done. A reply that ends in complete json fences, each a call to an offered tool, with only words before or between them, now runs them all (fenced_calls, ids react_N as other text calls); words after the last fence, other code, an unclosed fence or a tool that wasn't offered keep it text (#753)
  • A call followed by a summary isn't run, and the model is told to retry it without trailing text: in a second live run of #417's case both turns put "### Summary:" after their one json fence, nothing ran, and the build stopped after two turns with no files. The next message now says that it wasn't run — PR #1028, branch bugfix/r5.4-fenced-call-summary
  • Keep the retry condition when Action: or Final Answer: appears between fenced calls, without rejecting ordinary words between calls (PR #1031 review) — claimed: Codex, branch bugfix/r5.4-fenced-call-followup
  • A call to a tool that isn't offered comes back from a llama.cpp server as text in Qwen's markup (<tool_call><function=web_search><parameter=query>…), and the loop takes it as the final answer: a live build on a 512k-window Qwen model stopped at its second check in 3 minutes. Read that markup as a call when it names an offered tool; when it names one that isn't, say so and name the tools it has, and ask again — claimed: youkaicountry, branch bugfix/r5.4-text-tool-calls
  • A local builder spends its rounds on the project file and never writes the app: run 4 had a compiling Cargo.toml by check 3, its last, with main.rs still "Hello, world!". See whether more rounds get there, or whether a small model needs the window and the --screenshot option written for it (a template in the Rust steps)
  • The model ran cargo new app in app/ (commands run there), so the project landed in app/app/ and every check said "No project to build found in app/ yet". The check now says where a cargo or npm project one folder down is, and the Rust steps say app/Cargo.toml, not a folder inside app/ (a brief can't say cargo init: not every builder's commands start in app/). Live, with calls after a sentence run too, the model then made app/Cargo.toml and the build compiled (#751)
  • usage.json after a local build says about 4,200 input tokens and 6.5 s of work, while Ollama logged about six requests of 4,000-5,000 tokens each: find what it leaves out. Cached tokens: Ollama 0.40 reports them apart (cached_tokens), and the built-in loop's TurnDone left them out while coding agents count them. It counts them now; a live run on Ollama 0.40.7 had recorded 5,323 input tokens of 75,014 — claimed: SoldierPePe, branch bugfix/r5.4-usage-cached-tokens
  • Told that its cargo new app made app/app/, the model kept editing app/app/Cargo.toml and running cargo build in app/, and checks 2 and 3 still found the project one folder down (live, GPU). See whether naming the move gets it out, or whether the run should move it
  • The model reads file paths from the project folder and commands from app/ as one base: it ran cargo init app and cd app && cargo build after writing app/Cargo.toml. build_command_run's result now says where it ran (ran_in), a command that makes a project in app/app/ is told it goes in app/ itself, and one that starts with cd app where there's none is told to leave it out. Live (Ollama, qwen2.5-coder:14b) the first command was still cargo new app, before any result could say so. On a GPU (Ollama 0.40.7) the note then reached the model; on the CPU a second run wrote the call with a summary after it, so nothing ran (the open line on a call followed by a summary) — claimed: SoldierPePe, branch bugfix/r5.4-build-command-folder
  • Fall back to app/ when build_files_list, build_file_read, build_file_edit or build_file_write receives an app/-relative path from build_command_run output that does not exist at the workspace root (in_project_or_app) — claimed: d7do7e, branch bugfix/r5.4-build-tools-app-fallback
  • Normalize leading ./ components before checking the explicit app/ prefix in in_project_or_app so ./app/... cannot fall back into a nested app/app/ — claimed: d7do7e, branch bugfix/r5.4-build-tools-dot-app
  • An assistant turn with no text sends an empty string, not content: null, which Ollama refused as invalid message content type: <nil> (#403, #536)
  • Understand a tool call a model writes as JSON in its text (#417; implementation dfe9a4db2dcb3e27edbd67311cbb4202584e7af6). The fallback protocol reads only Action: and Action Input: lines, so anything else ends the turn with nothing done
  • Preserve length and content_filter stop reasons on native tool-call streams; a parsed call must not override truncation or refusal and execute tools. Pair other retained unexecuted calls with results before the conversation continues — PR #544
  • Return a rejected tool argument's JSON Pointer when known and guidance to correct the call (#498), keeping schema types strict — commit 0c3ae5cca569
  • Find how each local server is told its context size and set it, or stop with a plain message when the prompt won't fit. We assume 24k and send no num_ctx, which an OpenAI-style endpoint may ignore anyway; LM Studio refused a 12.6k prompt on its 8k default (#73). Ollama 0.40.15 loads a model with 4096 tokens unless told otherwise: its window is now read (/api/ps), so the budget is 3072, but a builder's prompt alone is larger
  • Read the context window from the server before the first request, so a large local model isn't trimmed at 24k: llama.cpp's /props, llama-swap's /upstream/<model>/… (its model list has a window only when its config gives one), Ollama's /api/ps, LM Studio's model list, and context_length in an OpenAI-style model list (OpenRouter). Only the window a server has loaded counts (not Ollama's /api/show, which says what the model was trained with), so Ollama and LM Studio say it once the model is loaded: asked again after the first turn. Keep 24k when the server says nothing, and a budget the person set stays the most — claimed: youkaicountry, branch feature/r5.4-read-server-window
  • Each AI's tool list is Auto, Lean or Full. Auto offers every tool when the window is 64k or more and the lean builder list (#417) below that; the memory steps in the system prompt follow the same choice. Today every local builder is lean, whatever its window. tools in [[ai.backends]]; Auto decides on the conversation's budget (FULL_TOOLS_BUDGET, a 64k window less its quarter) after the server is asked, and decides again when a later answer changes it (Ollama says only once loaded); paid services keep every tool. Only builders are trimmed: a chat keeps every tool whatever the setting — claimed: youkaicountry, branch feature/r5.4-builder-tool-list
  • Run a build live with the full list on a large local model (a 262k or 512k window), and note in STATUS which models used the extra Core tools well. A Qwen model with a 512k window: 2 of 3 full-list builds of "a scientific calculator" were Ready, using the brief, handoff and finding tools; the third stopped when the model wrote a call to a tool that doesn't exist as text
  • Run a build live with the lean list on a small local model, and note the result in STATUS
  • Each AI's tool list in Settings → AI (beside what Auto works out to) and ai tools <name> — claimed: justinjohn0306, branch feature/r5.4-tool-list-setting
  • A builder that can't see pictures (the API loop) is told it can't, then sent to open screenshots with Read: the website capture lines and the --screenshot and reference-picture lines in brief.rs, the website check's "Open each pair", the reminder after each round, the --screenshot contract and capture/README.md's window picture. Give it only what it can use, and say what the check measures instead. BriefCx::sees_pictures (today: not the API loop) picks the wording: a website's layouts and each round's capture.json, a program's ui-tree.json and the check's likeness, and asking the person what their pictures show — claimed: youkaicountry, branch bugfix/r5.4-blind-builder-prompts
  • The carry-on and follow-up prompts (continue_prompt, follow_up) ask an API-loop builder to check its work with rebuild_check, which the loop refuses while it builds; only its first message says so. Say "Craft Anything checks it after your turn" there instead — claimed: Afloat16, branch bugfix/r5.4-api-follow-up-check
  • When a server refuses a request for its context window and says the size (Ollama, llama.cpp, OpenAI-style), fit the conversation to that window and ask once more, else say how to raise it (#962)
  • Ollama fits a long prompt to its context by dropping the oldest messages, and a Qwen 3.8 model then refuses what's left with 500 "no user query found in messages" (#1245): say the context is too small and how to raise it, and don't retry what can't fit — claimed: justinjohn0306, branch bugfix/r5.4-ollama-dropped-request
  • Make room on its own after #1245's refusal: the server doesn't say its window, so this waits for the window read from the server (#1155)
  • A per-backend context budget in [[ai.backends]] (LoopConfig::context_tokens exists), for services whose real window differs from 150k (#374) — implementation: 6bd94b84828ec89b805cb8a246f557d0a778c037
  • A per-backend request rate in [[ai.backends]], for services that limit requests per second or minute (#668): requests_per_minute, shared by every conversation on the backend. Not run against a live provider — claimed: justinjohn0306, branch feature/r5.4-requests-per-minute
  • Offer requests_per_minute in Settings → AI and ai request-rate (#668) — PR #1841; claimed: rudycelekli, branch feature/r5.4-request-rate-setting
  • Say in the feed when a build is waiting for its turn (#668); a tokens-per-minute limit isn't covered either
  • Find LM Studio on another port (#73): the port in its settings file (when it is there) and the ports 1235-1239 are probed like the usual servers; OpenCode's own baseURLs were already read (#83)
  • Check live that LM Studio on --port 1235 is offered, and where its settings file is on Windows, macOS and Linux (LM_STUDIO_SETTINGS is from memory of its folders, not verified)
  • "Look for AI" (detect::probe_local) saved the first chat model a local server lists; LM Studio lists every model downloaded and FastFlowLM every model it can download, so it could load or download one the person isn't using (#1434). Prefer the loaded model: /api/ps (Ollama, FastFlowLM) or LM Studio's /api/v0/models (state: "loaded") — claimed: jgt87, branch bugfix/r5.4-detect-loaded-model — PR #1437
  • See why Bionic or LM Studio can't be picked (#140)
  • Do not select an embedding, ranking or transcription model as a chat backend when a local server has no chat models — claimed: Afloat16, branch bugfix/r5.4-nonchat-model-detection-20261103
  • Recognize specialised model file paths and numeric task-name versions during local chat setup — claimed: Afloat16, branch bugfix/r5.4-model-task-identifiers-20261103
  • Adding a server by its address with no model took the first chat model its /models lists; FastFlowLM lists every model it can download, so the saved model started a 5.5 GB download and unloaded the one in use (#1434). Prefer a chat model the server says is loaded (/api/ps, which Ollama and FastFlowLM answer) before the list — claimed: jgt87, branch bugfix/r5.4-add-server-loaded-model — PR #1435
  • decision When nothing says which model is loaded and a server lists more than one chat model, ask the person to pick one (the list in the address form's model field) instead of taking the first (#1434): /api/ps helps only while a model is loaded, and LM Studio and llama.cpp routers don't answer it
  • The first run's AI step can add a model server on another computer by its address (llama.cpp, Ollama or LM Studio on a home server), the same form as Settings → AI
  • A chat sent while a llama.cpp router is still loading the model ended after 31 s with "The answer stream ended early." and no next step (seen once live, 27B model); say the server may still be loading it and to try again in a minute — claimed: yashpatadiya1724-cmyk, branch bugfix/r5.4-stream-loading-remediation
  • An AI on this computer (Ollama, LM Studio) that can't be reached, or that took a request and didn't start answering, is told apart: "Couldn't reach Ollama (local)" for a server that was reached and "check your internet connection" for one on this computer misled, and an answer that never started didn't say a large model may still be loading or reading a long prompt (#1239 gave it the timeout_secs advice) — claimed: jgt87, branch bugfix/r5.4-local-server-messages — PR #1242
  • Adding a server by its address showed nothing while it connected and gave up after 10 s with "check the address" for a server that was reached but still starting or loading a model (#1429): show "Connecting to …" with a spinner and the seconds waited, wait a minute, and give advice for a server that was reached — claimed: jgt87, branch bugfix/r5.4-add-server-connecting — PR #1430
  • Let a person stop adding a server while it's asked for its models (#1429): the request runs on a helper thread and the wait checks the job's cancel, so Stop beside "Connecting to…" ends it at once (AiHub::add_server_until); the request given up on ends by its own limits — claimed: FuryNocturn, branch bugfix/r5.4-stop-adding-a-server

R5.5 API-key builds fail in words people can use

Owner: open · Status: todo

  • rebuild --backend <an API backend> still adds the first agent app it finds to the person's settings and prints "Using Claude Code (…)": ensure_agent (rebuild_cmd.rs) runs whenever no agent app is set up, though this run won't use one. Run it only when the build needs one — claimed: AlexJoaquimPereira, branch bugfix/r5.5-no-agent-with-api-backend
  • --backend <a name that is turned off or unknown> is refused with a specific message: a turned-off backend is named and must be turned on, and a name that matches nothing is named instead of saying to install an agent or add a key — claimed: AlexJoaquimPereira, branch bugfix/r5.5-disabled-backend-name; commit a380b3e
  • When a project's saved builder was turned off, Start builds with the default AI without saying so: the UI picker filters the stale name out (ui-egui/src/builders.rs) instead of reporting that the chosen AI was replaced. The Make picker says so in a toast, and every start (Start, Continue, a note, Try again) names the saved AI and the one building instead — claimed: AlexJoaquimPereira, branch bugfix/r5.5-stale-builder-notice
  • rebuild --continue-in, rebuild_start on a project and rebuild_question_ask name the saved builder that is off or gone and the AI building instead: the CLI prints it before the run, rebuild_start adds builder_notice, and an answered question carries it — claimed: AlexJoaquimPereira, branch bugfix/r5.5-headless-builder-notice
  • The saved-builder replacement notice is English only for now; add it to the catalogs after a native speaker checks the wording
  • Bound accumulated data across all lines of one server-sent event, so short lines cannot grow a streamed API response without limit — claimed: Afloat16, branch bugfix/r5.5-bounded-sse-events-20261103
  • Show the provider's own reason: #369 bought $10 of credit and got only "OpenRouter answered 400: Provider returned error". Find the reason in the response body and show it — PR #542
  • OpenRouter picks an upstream provider per request, and one without tool calling answered "400: Provider returned error" (#369): a request with tools now sets provider.require_parameters, and a 404 "no provider for these parameters" falls back to the text protocol like any server that rejects tools
  • OpenRouter reserves a model's whole output limit against the key's credit when no max_tokens is sent, so a key with $35 left of a daily limit got 402 "You requested up to 131072 tokens, but can only afford 35063" (#1213): ask again once with the answer capped at what it can afford — claimed: justinjohn0306, branch bugfix/r5.5-openrouter-afford-tokens
  • A saved fal.ai key reaches an API-key build's asset tools; every call said "No fal.ai key" (#923). Outside a project the tools now say that instead. Not checked with a real fal.ai key or API build
  • An API-key build works alone: it shares the app's session, whose attach points at whichever project started last, so another build beside it sent its files and asset spending to the wrong project (build.parallel is 2 by default). Others wait their turn in the queue
  • Let API-key builds run side by side: give each an engine of its own, as a question has. This also closes the gap left in working alone: the app and the rebuild_start tool each keep a queue, so a build the in-app chat starts isn't kept apart from one the app started
  • Check #369 live with Kimi K3 and a key limited to Kimi/Moonshot models (needs the person's OpenRouter key): that the build starts, or that the error now names the provider's reason
  • A live API-key build (needs a key on the test machine)
  • A Google AI Studio (Gemini API) key is its own choice in Settings → AI and first run (#646): the gemini backend kind uses Gemini's OpenAI-compatible endpoint and GEMINI_API_KEY
  • Check a Gemini build live with a real Google AI Studio key (tool calls through the OpenAI-compatible endpoint)
  • Failure logs for API-backend stops, like the ones agent CLIs write to logs/ in the config folder: a failed model turn writes one (agent::write_failure_log) and the error names it
  • Failure logs for an MCP server that doesn't start: a local [[plugins.mcp]] server that can't start, exits or doesn't answer writes one to logs/ (what ran, its exit code and stderr, never its env) and the error names it; an exit says what it printed. A remote server gets none
  • Review of #1921: a key in a plugin's arguments (?token=…) is hidden in its log, a server that stops answering is stopped and its error and log keep what it printed, and a failed write waits a moment for the exit. The wait after a failed write has no test of its own (it's a race)
  • An API backend's failure log leaves out what the service said: write_failure_log wrote the error's message, where a service or a proxy in front of it can repeat the key or words from the request (raised in review of #1049). It keeps Craft Anything's code, the HTTP status, the context window and request size, and the service's error code only when it is one of failure::KNOWN_PROVIDER_ERRORS — claimed: scks1, branch bugfix/r5.5-api-failure-log-service-text
  • Settings → AI can replace or forget an API backend's saved key, so a key rotated in the environment isn't hidden behind an older saved one (#1358); ai key <name> shows, replaces or forgets it headlessly, and never prints it — claimed: AlexJoaquimPereira, branch bugfix/r5.5-replace-saved-key
  • A tool that needs consent, refused by an API-model build without asking, told the model "The user said no" (#1938): it now says nobody was asked and points at a plugin's consent setting (Conversation::refuse_consent, BUILD_REFUSES_CONSENT) — claimed: justinjohn0306, branch bugfix/r5.5-build-consent-refusal
  • decision: should an API-model build ask the person (a card, as coding-agent builds do through the project's inbox) before a tool that needs consent, or follow a setting, instead of refusing every one outside the Build tools (#1938)

R5.6 Tools report only what they saw

Owner: open · Status: todo Golden rule 3: the result is evidence. Each fix needs a Mac to test on.

  • core_report_write renders only the first 30 distinct limitations in Markdown and HTML with no marker for the rest; say how many were omitted in both formats. — claimed: AlexJoaquimPereira, branch bugfix/r5.6-report-limitation-count
  • macOS: a denied Accessibility permission is reported as success with an empty tree (#240) — claimed: mavericzz, branch bugfix/r5.6-accessibility-denied; PR #887
  • macOS: runtime_ui_read captures a requested screenshot (#248): one on-screen window belonging to the pid, restricted by window_id when supplied; PNG/base64 up to 2048 px per side. Combined JSON (including escaping and the newline) stays within 16 MiB by trimming the tree with explicit truncation and partial coverage, preserving the image. Missing Screen Recording permission, a missing window or a capture failure produces an explicit screenshot limitation and partial evidence. Denied Accessibility permission with the tree enabled (the default) returns a tool error before capture; accessibility: false skips the tree. Requires macOS 14+ and existing Screen Recording permission; opens no dialog — claimed: Codex, branch bugfix/r5.6-macos-screenshot
  • macOS: runtime_ui_read honours accessibility: false (#307): the Swift helper skips the tree and its permission check, returns no nodes without truncation, and the evidence records the flag and says the tree was skipped. Walkthroughs still require Accessibility — claimed: Codex, branch bugfix/r5.6-macos-ui-flags
  • macOS: call observation's lldb callback was a literal \n in its Python, so the Python was invalid and it reported zero calls (#251). Fixed under R2.4: the callback is one valid line covering every breakpoint, and a failed lldb run is an error, not "0 calls"; only the live lldb run is still owed (R5.7) — claimed: mavericzz, branch docs/r5.6-stale-251; PR #1070

R5.7 Live checks owed

Owner: open · Status: todo The code for each of these is done; what's left is to run it for real and record the result in STATUS.md. A failure becomes a new line under its item. Grouped by what the run needs. Start with the Clean room build and the README quick start: Clean room is the default and has no live build yet, and the quick start is how every new person begins.

A signed-in coding agent:

  • A live Clean room build of a website and of an app, checked end to end
  • A live build in a picked folder each way (edit in place, start from), checked end to end
  • Thresholds under pressure: a clone that's off by a few pixels (a page with fluid type or clamp() spacing) to see whether 4 / 2 / 8 px is too strict
  • Screenshot check: in a live run the final check once got no picture from an Electron app within 90 s while the same check passed in the builder's turn and on a re-run; not reproduced. The app waited on animation frames, which a window behind others may not get; the Electron steps now say to capture after loading instead
  • A live Windows build whose check stalls (Codex, a Godot project, as in the reports): the feed names the check it is in, the builder's own checks stop at their 30-minute limit, Pause acts between checks and stops a running build step, and a caught bug leaves "Needs you"
  • A live chat run with a big comparison result, to see Claude Code page it with core_evidence_browse (and whether 80 KB is the right size). On macOS with chat --deep, Claude Code 2.1.322 and two arm64 builds of cargo (Rust 1.97.1 and 1.99.0, ~30 MB each), whose app_builds_line_up page is 581 KB: Claude Code asked for detail: summary itself, then paged with core_evidence_browse (matching, fields; 7–14 KB a page) and quoted the rows; no result reached the budget (largest 25 KB) in 23 calls ($1.14). Without detail, the server answered the 581 KB page summary-first (4.8 KB, continue_with). 80 KB holds: a 79.5 KB page (limit: 15) reached Claude Code whole as 37.7 KB of text, about 17k of its 25k-token tool limit, and it quoted the page's last row correctly — claimed: chordpli, branch test/r5.7-evidence-browse-live
  • Neither the line-up's summary nor core_evidence_browse's error names the lists it can page: Claude Code asked for list: "changed" (a show value), got "changed isn't a list of records… Leave them out to page it as it is", and ran the 20 s line-up again with show: changed before finding list: "pairs" with matching. Name the lists (pairs, ambiguous, only_left, only_right) in both. Every record's summary view (a detail: summary answer, the summary-first answer over a connection's budget, browse's kind: summary) now has lists, name → length, for the names browse reaches a list with; browse's errors for a name that isn't a list name them too — claimed: chordpli, branch bugfix/r5.7-browse-list-names
  • Live-check the named lists in a Claude Code chat: the same two cargo builds, and the agent pages pairs straight from the summary, without a wrong list or a second line-up. Claude Code 2.1.322, same question: the summary carried lists (pairs 1448, ambiguous 612, only_left 29746, only_right 32342); it ran the line-up once and every browse call asked for pairs, none failed (22 calls, $1.06) — claimed: chordpli, branch bugfix/r5.7-browse-list-names
  • app_builds_line_up paired 1,448 of ~36,000 functions across two Rust toolchains: every v0 crate hash (Cs<hash>_) changed, so exact names rarely match, and some masked-bytes pairs are unrelated functions (hex_encode → _git_oid_tostr). Compare Rust names with the crate hash left out. A new name_without_hash stage compares demangled names with the instantiating crate kept, and calls are compared the same way: the same two cargo builds now pair 14,036 (13,356 by name), with masked-bytes pairs down from 888 to 233 — claimed: chordpli, branch bugfix/r5.7-rust-hash-free-names
  • After that, 1,347 of the 1,630 name groups left ambiguous hold one symbol at several addresses in one build (a local copy per codegen unit): pair copies whose bytes agree, or say they're copies. hex_encode → _git_oid_tostr is now a low-confidence structure pair (the new build inlined hex_encode)
  • Live check on Windows: rebuild a project while its frozen pin runs, and the build isn't blocked

An API key or a bill:

  • Run Nano Banana Pro once with a real key (a picture, then a 3D model from it) and check the picture's size and the invoice ($0.15)
  • fal bills pictures "by rounding up to the nearest megapixel"; estimates use the exact size. Confirm on a real invoice whether a 1024×1024 picture bills as 1 or 2 megapixels, then round in the table (needs a key)
  • A live extraction run with Kimi K3.1 Pro (needs the person's OpenRouter key)

A second computer or another system:

  • A live run of a project copied between two real computers with Claude Code (#451, #452); done on one Mac with a second CLAUDE_CONFIG_DIR standing in for the other computer (STATUS.md)
  • Ghidra setup tried for real on Linux: the guided install (Ghidra 12.1.4 and its Java) on a clean Debian 13 and a Debian 13 desktop, then a Deep-mode decompile from Ghidra (STATUS.md, Clean room and Deep mode) — claimed: Fadlalla-ai, branch test/ghidra-setup-linux
  • Ghidra setup tried for real on macOS (the Java pins were checked by tests only)
  • The Java that Ghidra's guided setup brings (<store>/jdk) isn't found as the Java runtime engine, which is looked for in <store>/java: engines_list still says Java isn't installed after the setup that its install message points to, and tools that need Java (Vineflower) don't use it. A guided install now names its store folder in the engine table (Install::Guided::folder), and the lookup and installs use it. Live in Debian 13 after engines_install ghidra: Vineflower decompiled a jar with that Java (before: "Java runtime is not installed") — claimed: Fadlalla-ai, branch bugfix/r5.7-managed-java-found
  • A path set in Settings → Engines (engines.tools.<id>) that names a folder, such as a JDK's folder for the Java runtime, counts as found: engines_list, doctor and Requires::Engine say it's there, and running it then fails with "permission denied". Look for the program in the folder, as an install folder holds it, and say so plainly when it isn't there. The program is now looked for in the folder; one without it isn't found, and the message names the folder. Live in Debian 13 with Java set to a JDK's folder: Vineflower decompiled (before: "permission denied") — claimed: Fadlalla-ai, branch bugfix/r5.7-engine-path-folder
  • gdb, rizin sessions, adb, JADX, Apktool, Frida, tshark, lldb and the browser look up their program with engines::locate and their own not-found message, so a folder set as their path without the program says it isn't installed rather than that the folder has no such program; use Settings::engine_not_found as the Java and disassembler tools do
  • A Java set for Ghidra (Settings → Engines, engines.ghidra.java_home) isn't used as the Java runtime either, so Vineflower still says Java isn't installed with only that set
  • JADX doesn't use the Java runtime engine (the Java Ghidra's setup brings, or a Java path set in Settings → Engines): its launcher script finds Java only through JAVA_HOME or PATH, and neither is passed, so with no system Java, Java source of an Android class fails although engines_list and doctor say the Java runtime runs JADX. That Java is now passed as JAVA_HOME when <home>/bin/java is there. Live in Debian 13 with no system Java, after Ghidra's setup's Java and engines_install jadx: Java source of a class came back (before: JADX exited 1) — claimed: Fadlalla-ai, branch bugfix/r5.7-jadx-uses-managed-java
  • A Ghidra folder setting that names a file, such as the /usr/bin/ghidra start script, is listed as found by engines_list, but Deep mode then quietly decompiles with the built-in lifter instead; refuse it, or follow the script to its Ghidra folder. A file is now run only when it is the headless launcher or a script that runs it; a link or start script is followed to its Ghidra folder; anything else isn't listed as found. Live in Debian 13 with Ghidra 12.1.4 laid out as Arch-based packages do: /usr/bin/ghidra and a linked ghidraRun now decompile with Ghidra — claimed: Fadlalla-ai, branch bugfix/r5.7-ghidra-setting-names-a-file
  • Two cases left by the line above (review of #1949): a wrapper script in Ghidra's own folder that runs the launcher was passed over for that folder's launcher, and a start script naming a link onto PATH (exec /usr/bin/ghidraRun) wasn't followed through it — claimed: Fadlalla-ai, branch bugfix/r5.7-ghidra-wrapper-and-linked-script
  • Settings → Engines looks Ghidra up on every frame it shows "Installed" (engine_ui.rs, the hover text calls ghidra_launcher): folders searched, and a start script read, while drawing. Keep the path found with GhidraFlow::ready instead. The launcher's path is now kept where "found or not" was (GhidraFlow::found), and looked up again after a settings change or install — claimed: Fadlalla-ai, branch bugfix/r5.7-settings-ghidra-lookup-once
  • A Ghidra with no decompiler program for this computer (public downloads leave out macOS) is refused before the long analysis, with the command that builds it; checked on a real Mac
  • engines_install builds Ghidra's native programs on macOS, so a Mac install works unaided
  • Follow the README quick start on a clean machine (Windows)
  • README quick start on a clean machine (Linux), install half: clean Ubuntu 24.04 and Debian 13 needed build-essential, now in the README's package line; then the app and CLI built and doctor was healthy (STATUS.md, "Linux quick start") — claimed: Fadlalla-ai, branch test/r5.7-linux-quick-start
  • README quick start on a clean machine (Linux), the window and the snake rebuild: the window opened on a clean Ubuntu 24.04 (X11, virtual screen) once libxkbcommon-x11-0 was installed (now in the package line), and the snake-game rebuild reached Ready with Claude Code on a desktop that wasn't clean (STATUS.md, "Linux quick start") — claimed: Fadlalla-ai, branch test/r5.7-linux-window-and-rebuild
  • README quick start on a clean machine (Linux), still open: the website rebuild examples, a rebuild on a clean machine, the window on a real screen, and a Wayland session
  • A Rust build whose folder's Cargo settings move its output (build.target-dir or CARGO_TARGET_DIR): built_program reads CARGO_TARGET_DIR, the closest project or ancestor config/config.toml setting, then the default target/, so Play and the checks find the program. A target directory outside the app supplies only a program the manifests name; without a name, only target directories inside the app are searched, and a pin copies such a program into its own folder; commits 842f011a, 216295fd — claimed: AlexJoaquimPereira, branch bugfix/r5.7-cargo-target-dir
  • Ask Cargo where a Rust build went where its program starts or is checked (Play, pins, the build's run and screenshot checks): launch_now runs cargo metadata and looks in its target_directory first, which covers $CARGO_HOME/config.toml, CARGO_BUILD_TARGET_DIR and a parent workspace's target/; the Project screen keeps the file lookup. Checked on a workspace member and a CARGO_BUILD_TARGET_DIR build that the file lookup doesn't find — claimed: Fadlalla-ai, branch fix/cargo-target-dir-at-start
  • With CARGO_TARGET_DIR set (a build folder shared by several projects), a program of the same name built by another project there is taken for the project's own: Play and a new pin can start or copy it, and a pin made before programs were recorded looks outside its copy. Take a program from a target directory outside the app only when Cargo's record of it (<name>.d) names the app's sources, and start an older pin from its own copy only. Done: the newest of the project's own builds is started; an older pin looks only inside its copy — claimed: Fadlalla-ai, branch bugfix/r5.7-pin-own-program
  • Build the new DMG on a Mac (release dry run) and look at the window once
  • #490 on macOS: a real Deep build through Claude Code with the published 0.3.0 CLI, with 2.1.179 and a current version. Windows passed; the Mac CLI was only checked to contain the fixed schema
  • macOS CLI zip blocked by Gatekeeper when downloaded in a browser (#94), though Apple has its notarization tickets: find why the online lookup fails for people (it passes on our Mac, which may be exempt), on a machine that shows it

A real program, device or file:

  • Cutter sessions: try them against a real Cutter, including whether =h& keeps Cutter responsive (only a stand-in server so far)
  • Vineflower: run it once on a real JAR (no Java on the machine it was written on) — claimed: hasnain607, branch docs/r5.7-vineflower-live-run
  • Verify the 23 unverified adb tools; record results in tool-status.md
  • Android locales: try only_locales on a real APK (only the test fixture so far)
  • Ask LM Studio and a plain llama.cpp server for their context window (R5.4, crate::window): llama-swap (its list and its /upstream/ route) and Ollama 0.40.15 were asked live; these two probes rest on their documented answers
  • app_graphs_compare on the two real Electron releases again, to count how many of the 162 "removed" now pair (and pair the index-<hash> chunks by structure if many stay)
  • A real IL2CPP global-metadata.dat (needs a Unity build or a shipped game); synthetic so far. Compare a real APK's components against another tool's reading
  • Stand-in backend: a browser-made HAR and a real front end talking to it (only a hand-written HAR so far); GraphQL and tRPC on real bundles (only hand-written minified shapes)
  • Check the Kimi Code, Factory Droid and WorkBuddy registrations in the real apps

M1 Extensibility

R1.5 External engine trait in L3

Owner: open · Status: in progress

  • Find Chrome, Brave, Chromium and Edge installed as Flatpaks on Linux (system and per-user exports), and keep a Flatpak or Snap browser's private profile in the home folder its sandbox shares, since its /tmp is its own (#747). Unit-tested; not run on a real Flatpak
  • Find browsers under their other Linux names, and say why one isn't found: a set path that isn't there, or a Flatpak's sandbox (#841). Unit-tested; not run on Linux or in a real Flatpak
  • decision Let the Flatpak start browsers on the computer. It can't see system ones today; a per-user Flatpak browser in ~/.local/share/flatpak/exports/bin is found, but its start script needs flatpak, which the sandbox doesn't have. flatpak-spawn --host would need --talk-name=org.freedesktop.Flatpak, which lets the app run anything outside its sandbox (#841)
  • Say in the feed while a decompile waits on Ghidra, and for how long, as the one-time analysis does (#1714): once a minute, "Ghidra is decompiling this function (N min so far; it gives up at 15 min)", or that it waits for another decompile to finish. The builder's tool server writes the line to inbox/activity/, which the run reads into the feed. Unit-tested; not run with Ghidra — claimed: justinjohn0306, branch feature/r1.5-decompile-wait-in-feed
  • Bound unguarded jump tables when Ghidra decompiles a function (#1715): with no bounds check it reads up to 256 entries, follows targets inside other functions and the decompile grows without end. Before a decompile the bridge finds a byte-indexed table with no guard, decodes the function's stretch (its .pdata entry on Windows) in a row, ends the table where another starts or at the first entry that isn't an instruction start there, and writes the override. Live on Ghidra 12.1.4: a Rust async function that timed out after 15 min decompiles in 16 s (121 tables) — claimed: justinjohn0306, branch bugfix/r1.5-bound-unguarded-jump-tables
  • The jump-table sweep finds the next function with a loop that always breaks on its first pass, which fails SonarCloud's reliability gate on main (java:S1751): take the first function from the iterator instead — claimed: justinjohn0306, branch bugfix/r1.5-ghidra-sweep-next-function
  • Stop Ghidra's one-time analysis when the project whose build started it is deleted (#948): it carried on, with a core and gigabytes of memory, until the app closed. Pausing still leaves it running, so a paused build loses nothing — claimed: justinjohn0306, branch bugfix/r1.5-stop-analysis-for-deleted-project
  • More pins once their hashes are checked against the publisher: Chromium (no published checksums found yet), Rizin for Linux (the tar.xz unpacks to ~800 MB, so it needs streaming extraction) and macOS (a .pkg)

R1.6 Headless first

Owner: maintainers · Status: in progress

  • Give Linux-specific Clang guidance in Unreal's guided installer when the bundled compiler is missing (the requirements summary is already platform-specific). — claimed: AlexJoaquimPereira, branch bugfix/r1.6-linux-unreal-clang
  • Give macOS the right C++ guidance when Xcode is missing: the status line and requirements summary now name Xcode, matching the guided step.
  • Find Visual Studio's C++ tools on Windows when Visual Studio is installed outside Program Files (through the Visual Studio Installer's ProgramData instance state or developer environment variables, #669, PR #726) — claimed: riyaz, branch bugfix/r1.6-windows-vs-cpp-tools
  • Turn off web and capture tools by default in restricted MCP mode (#856, #855).
  • Find a Unity Editor that was added to Unity Hub with "Locate" (#1077): read the Hub's own list of editors (editors-v2.json, editors.json), on macOS and Linux too. Unit-tested with both layouts of the list — claimed: justinjohn0306, branch bugfix/r1.6-unity-located-editors
  • Check on a computer with Unity Hub that a located editor is found (#1077): the list's layout was taken from descriptions of it, not from a real file

M2 Product completeness

R2.2 More coding agents

Owner: open · Status: in progress

  • Insert a missing YAML server table before an explicit document end marker. — claimed: Plastic-time, branch rrfz/bugfix-r2.2-yaml-document-end
  • Find Codex bundled with ChatGPT.app/Codex.app after a macOS Dock launch, and verify the selected CLI's sign-in through the headless connection check and Settings. — claimed: dbzx6r, branch bugfix/r2.2-macos-codex-signin-upstream
  • A saved agent program path that no longer exists (Codex desktop moves its CLI to a new build folder on every update) says so and names the path, instead of "isn't installed or isn't on your PATH" (#721); it stays a missing program, so a builder turn isn't retried
  • A relative program path started in another folder is looked up where it was started before it's reported as missing (Unix: a script whose interpreter is missing was said to have no file there)
  • ai test names the program it ran and the saved one it replaced, and doctor says a saved path has no program there instead of blaming PATH (#721)
  • doctor says "Several copies are on PATH" for one file reached twice, when a PATH folder is a symlink to another (/bin -> /usr/bin, so both spell the same program, #1550): compare where the copies really are before counting them — claimed: muraa-p, branch bugfix/1550-doctor-symlink
  • doctor looks a registration's bare command (craft-anything-cli, as a hand-made entry names it) up on PATH, as the app starting it does, instead of calling it missing. — claimed: Plastic-time, branch bugfix/r2.2-doctor-cli-on-path
  • Settings says when a saved agent program is gone and which one runs in its place (AiHub::replaced_programs, read on a worker with the AI list), with "Use it" to save that one when it's a file of its own (#721, #1468) — claimed: FuryNocturn, branch feature/r2.2-saved-program-gone
  • A build says in its feed when it runs another program than the one saved in Settings → AI (#721) — claimed: Agapenordev, branch bugfix/r2.2-build-says-replaced-program
  • GitHub Copilot through its own CLI login and ACP, without credential import: discovery, MCP-only tool restrictions, model selection and fresh read-only questions; unit tests and CLI 1.0.105 with an isolated home and loopback model, all five access modes — claimed: qu-gg, branch feature/r2.2-copilot-integration (recorded during PR preparation)
  • Grok Build as a coding agent over ACP (grok agent stdio): discovery in ~/.grok/bin, the person's own grok login (a missing auth.json is unknown, so an API key still starts), model and reasoning effort, and our MCP server on the session with "type": "stdio" only for Grok. Clean room, deep chat, and questions are refused, because Grok can read and edit without a permission request. Verified against Grok 1.0.77's ACP handshake and session/set_config_option (the CLI --model flag does not stick after session/new). A live deep build on 2026-11-06 (Grok 1.0.77) finished, called the Craft Anything server, and answered one permission request. Clean room and deep chat were not run live; those runs are refused. See Grok Build. — branch feature/r2.2-grok-agent
  • Review hardening: cap target handoff at 128 while preserving the active target; propagate context-discovery cancellation, normalize CLI-only efforts for Claude requests, and name the configured agent in shared picker warnings, with focused regressions and translations
  • Local agent chats share a saved investigation and open-target metadata with MCP; background refresh makes recorded findings visible in the workbench Evidence pane. core_finding_edit corrects a finding with checked evidence and an optional revision guard, preserving confidence when omitted. Builder refresh failures warn instead of ending the turn; shared ACP errors use the current agent and its sign-in guidance. Engine, process-level MCP and headless rendering tests cover sharing, mode boundaries and finding ids; regressions cover confidence preservation, explicit overrides and agent-specific errors
  • Serialize project-backed investigation load, revision check, mutation and all rendered writes under one bounded, cancellable interprocess lock. Refuse malformed state rather than replacing it; separate-process regressions retain concurrent appends and reject one of two updates based on the same revision
  • Refresh externally changed investigation/evidence records incrementally, avoiding repeated full reloads under the session lock while retaining mid-turn findings visibility
  • Offer OpenCode locally and over SSH with its own visible chat models and supported thinking depths (Low, Medium, High, Xhigh, Max) in settings and headless clients. Choices are validated before saving; explicit choices are refused at execution until native MCP support is available. See OpenCode.
  • Capture OpenCode settings and project together; refuse a busy session instead of treating it as no project, and preserve open model pickers and matching results during contention.
  • Report cancelled model validation as stopped, while preserving error events for other failures.
  • Finish rejected or cancelled model validation with a terminal event so headless chat exits; document functions touched by the OpenCode model-selection changes.
  • Integrate the five-level Effort type from #395 for agent variants, and share a strict disk-based selection save between the hub and headless tools, preserving unrelated edits.
  • Rebase onto main after #395's squash merge, preserve extra model arguments without a saved selection, and render the remote-host placeholder visibly in the setup instructions.
  • Hold the UI settings worker's session guard through its disk save so it cannot overwrite a newer model selection; reverify the existing terminal-event regression for validation failures.
  • Label the settings-save guard so filesystem I/O is identifiable as a running job by the lock tracker.
  • Finish a failed CLI worker-thread start with a terminal event after its error so headless chat stops polling.
  • Synchronize the UI exit-time settings save with the session's latest AI table and other writers; an isolated exit-save regression preserves the latest model/depth and pending app settings.
  • Never wait for a running tool's session lock at shutdown: defer non-AI app edits in a versioned sidecar when busy, then recover against the latest committed AI settings at startup.
  • Discard superseded deferred settings after a successful full exit save; document model-only and model-with-variant command forms, and share transport selection between preflight and runs.
  • Rebase after #567 and translate the ten model-selection strings and two agent tool titles in the Russian catalogue, restoring coverage and catalogue key parity.
  • Supply and verify the run's Craft Anything MCP server in native OpenCode before enabling explicit model/thinking execution: a config file in OPENCODE_CONFIG with run --standalone (verified live locally). — claimed: advokat11, branch bugfix/r2.2-opencode-native-mcp
  • Hand native OpenCode runs over SSH the Craft Anything MCP server (the config file is local). One ssh run checks craft-anything-cli is on the host's PATH and writes the config on the host (~/.cache/craft-anything/, mode 600, atomic rename); the builder fails fast with the fix if the helper is missing. Verified: POSIX-shell regression test (the_remote_line_runs_on_a_posix_shell) and an SSH round-trip to localhost (probe, byte-exact write, mode, cleanup). Not run with a live agent over SSH. — claimed: jelassiaymen94, branch feature/r2.2-opencode-remote-mcp
  • Make the_remote_line_runs_on_a_posix_shell accept the shell's missing-command status: assert failure rather than 127, isolate PATH from installed helpers, and verify that a missing helper writes nothing in a fresh home. Fixture utilities use host lookup rather than assuming /bin; the shell uses the bounded process runner. Reproduced the old failure on macOS; the AI suite now passes (446 passed, 2 ignored). PR #1018 — claimed: Codex, branch test/r2.2-posix-remote-helper
  • Find OpenCode's CLI inside its Windows desktop app (opencode-cli.exe), for detection and for runs (#1060)
  • OpenCode Go gets one session id per installation, saved at startup for any saved Go backend, so a restart keeps it (#1225). An OpenCode error without a message shows its name, status and the start of its response body. Not run live; the run-5 cause in #1225 still needs its log. — claimed: fabox-os, branch bugfix/1225-opencode-go-session
  • Settings → AI can only add an agent it detected: add one by choosing its program, for a CLI kept somewhere we don't look (#1060)
  • Verify OpenCode model discovery and an explicit thinking selection on a live SSH host, and complete a live generation with the selected model and variant (local catalogue verified).
  • A stopped, failed or killed OpenCode turn deletes its session, so OpenCode's shared service can't resume it and keep running the builder's commands later (#1115). Probed live with OpenCode 2.2.5 on Windows: a killed run --standalone session stays running and the service resumes it; session delete --standalone removes it without resuming it — claimed: justinjohn0306, branch bugfix/r2.2-opencode-interrupted-sessions
  • When Craft Anything itself is killed mid-turn (uninstalled, a crash; closing the app during a build isn't known to wait for the delete either), no turn is left to delete the OpenCode session, and the service still resumes it (#1115): each turn records its session under a lock the system drops when the process ends; the next agent turn and craft-anything-cli uninstall delete the sessions of records nobody holds. Over SSH, the delete runs on the host too — claimed: justinjohn0306, branch bugfix/r2.2-opencode-leftover-sessions
  • Sweep leftover agent sessions when the desktop app starts, not only at the next agent turn or craft-anything-cli uninstall (#1115) — a background thread at start-up, so a delete over SSH can't hold up the window — claimed: AlexJoaquimPereira, branch bugfix/r2.2-startup-session-sweep
  • A turn stopped before OpenCode's first event leaves no session id to delete: OpenCode names its session only in that event, so the session it may already have made stays (#1115, review on #1176). Needs a session id known before the run (one we create and pass in), if OpenCode allows it
  • OpenCode's model discovery (api model.list, run before every turn to check the chosen model) has no --standalone, so it starts the shared service, and the service resumes any unfinished session on the computer, in any project (#1115). Seen live with 2.2.5 on Linux: a build's model check woke it, and a session left from days before went on editing another project. Discover with a private server unless the person chose --server — claimed: youkaicountry, branch bugfix/r2.2-opencode-discovery-standalone
  • OpenCode's private server can answer model discovery with 503 ServiceUnavailableError ("Location plugins did not finish activating") while it starts, now and then, and the question or build stops before it begins (#1859). Ask once more after a short pause when that is what it said; a turn already retries OpenCode's 503s — claimed: FuryNocturn, branch bugfix/r2.2-opencode-models-503
  • OpenCode prints that typed answer on stdout and HTTP 503 Service Unavailable on stderr, and a failed run's error keeps only stderr, so the retry above never sees the answer and the person is shown "HTTP 503" instead of OpenCode's reason (#1859). Read the typed answer from stdout, retry on it, and name its message in the error — claimed: justinjohn0306, branch bugfix/r2.2-opencode-503-stdout
  • docs/opencode.md says "OpenCode V2" but not where to get it: OpenCode's install script and npm's opencode-ai give the 1.x line, whose run has no --standalone (npm's @opencode/cli is 2.x), so the check before a build can't confirm it. Say where 2.x comes from and how to point the backend at it, and record the Linux live runs in STATUS.md — claimed: youkaicountry, branch docs/r2.2-opencode-live-findings — PR #1241
  • OpenCode's shared service, started while a --standalone run is going (opening the desktop app does), takes over that run's session: both servers ran its commands, and the turn ended with an error that had no details (seen live on Linux, a 2.2.6 service beside a 2.2.5 run). Keep a run's session out of the service's reach, if OpenCode allows it; until then docs/opencode.md says not to open OpenCode during a build
  • OpenCode logs "Received a progress notification for an unknown token" for our tool server after calls through 2.x's execute (seen live on Linux, 2.2.5): check that our MCP server sends progress only for a request that carried a progressToken, and only before its result. The server now accepts only a string or number token, and sends the notification before the response; a missing, null, object, array or boolean token gets none. — claimed: AlexJoaquimPereira, branch bugfix/r2.2-mcp-progress-token
  • Recheck the OpenCode 2.x "unknown token" log against a build with the progress-token filter; the live warning was not reproduced, and the server-side path is covered by tests.
  • Stop with a plain error when the opencode found is 1.x (its run --help has no --standalone), instead of running it anyway and showing its usage (#1111) — claimed: Agapenordev, branch bugfix/r2.2-opencode-v1-refusal
  • When the opencode on PATH is 1.x and an OpenCode 2 is installed elsewhere, find and use the 2.x one, and confirm a live OpenCode 2 build end to end (#1111 stays open until then)

"Found on this computer" has to mean "can do the job". Today only Claude Code gets everything. Every builder gets the same tools:

  • A table of what each adapter gets (our tool server, approvals, resume, notes mid-turn, a message of any length), as typed data, with a test that fails when one is left out. AgentAdapter already answers some, such as resumes and prompt_on_stdin. The lines below fill it in. cli_agents::grants states the columns an adapter can't answer (our tool server, approvals, notes mid-turn, and whether the run is per message or ACP), reports resume and a long message from the adapter itself, and four tests hold every row to what the adapter actually hands a builder: a run's arguments and environment (or the file an argument names, for the harness's overlay) must carry the server it claims, the notes hook must be in the arguments exactly when a note mid-turn is claimed, and a claimed approval channel must be there. A cell nobody has checked says NotYet with the gap in words rather than guessing; today that is DeepSeek Harness's approvals and every agent with no adapter of its own — claimed: muraa-p, branch feature/r2.2-adapter-grants
  • OpenCode's plain adapter never passes our tool server, so its builder has none of our tools and hunts for a setting that doesn't exist (#481). Hand it the server, or build over ACP by default once that passes
  • Aider has no adapter, so the plain one hands it the prompt as a file name, and it can open a browser sign-in nobody agreed to (#72). Give it an adapter, or stop offering it — claimed: Afloat16, branch bugfix/r2.2-aider-recommendations-20261103
  • Add a non-interactive Aider adapter before recommending it again; detection currently leaves it out. Aider gets the message with --message, and flags that stop it asking, updating, editing .gitignore or opening a page; a chat outside a build also gets --dry-run. Before a run it refuses when no model or key is set up, or an OpenRouter model has no OpenRouter key: both would offer a sign-in and wait for a browser nobody is at (#72). — claimed: hasnain607, branch feature/r2.2-aider-adapter
  • Aider: a Clean room run is refused in validate_access, but only the ACP path called it, so a native run (run_cli) still started Aider and its unchecked reads. Call it before the setup check, with a test through the launch path — claimed: hasnain607, branch bugfix/r2.2-aider-clean-room-run-cli
  • Codex (#499): in a chat the tool server has no active target. The reporter attached a patch for chat continuity; look at that first
  • Codex (#538): list every permitted tool from startup, since Codex doesn't refresh its callable list after enable_toolset changes it — PR #805
  • Codex (#540): rebuild_start was refused by codex exec's approval policy and showed as null; setup and the builder now approve our tools for Codex (we still ask before running code) and a refusal explains the fix (docs/codex.md)
  • Codex (#540): run codex exec on a real install and check rebuild_start gets through
  • Codex (#1867): a refusal of one of our tools that setup doesn't approve (build_command_run) tells the person to run setup --client codex, which changes nothing for it. Say how to approve that tool in Codex's own config.toml instead, and keep the setup hint for the tools setup approves (AGENT_APP_PREAPPROVED_TOOLS) — claimed: FuryNocturn, branch bugfix/r2.2-codex-blocked-tool-hint
  • Codex's approval guidance now adds the tool inside the inline tools table setup writes; the dotted-table form failed with "cannot extend value of type inline table with a dotted key" (#1867). A config-shape regression keeps the hint and the generated table valid. — claimed: AlexJoaquimPereira, branch bugfix/r2.2-codex-approval-hint
  • Codex: resume support (codex exec resume <id>), verified on a real install — claimed: Codex, branch feature/r2.2-codex-resume (local workspace) — verified on macOS with Codex 0.169.0-alpha.11.1: MCP call, sandboxed file write, follow-up and reopened thread through codex_resume_smoke; not a full rebuild — missing-thread stderr clears the unavailable session; parser and runner recovery regressions cover it. Resume argument order on older Codex releases is unverified — refuse --ephemeral before startup; two-turn regression checks that no session is saved
  • Aider: move the message to --message-file once the launcher can write one (see the argv item above). A local run writes the message (preamble first) to a private file in the temp folder, passes --message-file, and removes the file when the run is over; a file that can't be written falls back to --message (reads_message_file, ArgsCx::message_file) — claimed: justinjohn0306, branch feature/r2.2-aider-message-file
  • Aider: a --message saved in the backend's extra arguments replaced the request once it went in --message-file, since Aider runs --message first (review of #1883): leave --message, --msg (or --ms) and -m out of the saved arguments when the message is in a file — claimed: justinjohn0306, branch bugfix/r2.2-aider-saved-message
  • Aider: a saved -vm "old task" still replaced the request in its file, since -v and -h take no value and may come before m in one group (review of #1884): drop the m and its message from such a group and keep the other flags — claimed: justinjohn0306, branch bugfix/r2.2-aider-grouped-message-flag
  • Aider: on Python 3.10 a saved -v=m "old task" is -v -m "old task" and still replaced the request in its file (review of #1895): read an = after one short flag before m — claimed: justinjohn0306, branch bugfix/r2.2-aider-equals-m
  • Aider: read project-folder .aider.conf.yml and .env in check_setup, which needs the working directory passed into check_native_options. It reads the run's folder, the top of its git repository, then the home folder, as Aider does — PR #1210; claimed: rmsff, branch feature/r2.2-aider-project-settings
  • Check Aider's settings on the SSH host before its turn, including the remote environment, ~/.env, ~/.aider.conf.yml and ~/.aider/oauth-keys.env. Only a setup status returns. — claimed: AlexJoaquimPereira, branch bugfix/r2.2-aider-ssh-settings
  • OpenCode takes its message on stdin (opencode run with no message argument reads it there, V1 and V2), so a long one can't pass Windows' command-line limit (#402), as Claude Code, Codex, Gemini CLI and DeepSeek Harness already do — claimed: FuryNocturn, branch feature/r2.2-message-off-argv
  • Hermes, Maki, Devin and Antigravity still take the message in argv (prompt_on_stdin), and Aider still does over SSH, so a long one can pass Windows' command-line limit (#402). Aider over SSH needs the file written on the remote host. Hermes' -z takes its message only in argv; hermes chat -Q --query-file - reads it from stdin, but that is the chat path, not -z's one-shot run, so check what it prints and approves before moving to it. For the others, find in their source whether they read a message from stdin or a file
  • A note sent during a build reaches the builder at its next step on other coding agents with tool hooks, and on the API-backend loop between tool calls (Claude Code only today)
  • Session copies for other agents that keep sessions (ACP agents' session/load) (#451)
  • Hermes Agent's native Windows install (%LOCALAPPDATA%\hermes\bin) is searched, so an app started before the install, or by a launcher, finds it (#116)
  • mcp_servers: null (or ~) in a Hermes config.yaml takes our entry like an empty table, instead of being refused as flow style. — claimed: Plastic-time, branch bugfix/r2.2-yaml-null-table
  • Live-check #116 on Windows: install Hermes natively, start the app from a launcher, and see Settings → AI offer it; #116's "error after every message" has no text yet (asked)
  • Refuse multi-document YAML configs before planning edits or removing registrations. — claimed: Plastic-time, branch rrfz/bugfix-r2.2-yaml-single-document
  • A Hermes config.yaml that starts with a byte-order mark gets our entry in its own mcp_servers table, not in a second one that hides the person's servers from Hermes. — claimed: Plastic-time, branch bugfix/r2.2-yaml-config-bom
  • See whether Hermes still fails on 0.3.0 (#460: the fix its fixed label names was already in 0.3.0 when it was filed, and it gives no version)

Approvals:

  • decision One approval behaviour across agents, and whether we want an auto-approve mode apart from Clean room and Deep (Auto mode only governs what's asked through us). Today it differs by agent: Claude Code asks our approval tool; OpenCode's builder runs with --auto; Hermes's -z turns its approvals off on every run, chats included. Clean room's approval rule doesn't reach an agent that never asks us
  • Clean room outside builds (#564): a chat or question with Claude Code refuses the agent's own file, shell and web tools outright (--disallowedTools), not through its defaults; an ACP chat refuses them through an approver instead of a consent card (docs/claude-code.md)
  • Say on each adapter, as typed data, what it can be held to (asks us, its own sandbox, nothing), in a build, a chat and a question, and have runs act on that, once the decision above is made. Today Codex, Gemini CLI, OpenCode, Antigravity, Hermes and Maki are not held to Clean room in a chat (their own sandbox, if any, is not established)
  • A Deep build through Claude Code can start subagents again: --tools left out Agent when #637 trimmed the list (#1105) — claimed: justinjohn0306, branch bugfix/r2.2-builder-subagents
  • Check live that a Clean room chat with Claude Code can't read a file outside its folder, and that an ACP chat in Clean room still reaches our tools (its permission requests for them are allowed, only its own read/write/command/fetch are refused)
  • Check live what Hermes over ACP asks about: whether its file reads reach our approval policy
  • Maki's plain maki -p builder doesn't ask us before commands, and in Clean room its reads aren't checked (as with every agent that never asks us): say so in the approval decision above, or run Maki over ACP by default once that passes. Outside a build, tools the person allowed in Maki's own permissions (an MCP server's, or all of them with always_yolo) still run; the adapter can't see that config
  • Maki over ACP writes inside its working folder without a permission request (its built-in rule), and that folder is a build's project folder, so writes to our own records (inbox, manifest) skip the approval policy; maki acp takes no tool flags to stop it (0.7.0). Start it in app/ (paths in the prompt relative to that), or ask Maki for a way to make those writes ask; part of the approval decision above
  • OpenCode's own write tool reaches our records too: a builder wrote its progress notes to inbox/build-progress.json and stopped the build (R5.3 now names the records in the brief and says how to go on). Stop those writes in OpenCode itself, with its permission settings or by starting it in app/, and check it live — claimed: youkaicountry, branch bugfix/r2.2-opencode-records-fence. A build's config carries permissions rules denying its edit tool each name in workspace::OWN_NAMES but checks/; checked live with OpenCode 2.2.6 on Linux (its shell isn't covered)
  • An OpenCode builder isn't kept out of folders outside the project: OpenCode allows external_directory by default, and --auto approves what it would ask. Seen on Linux with 2.2.6: a builder kept scratch work in /tmp/opencode, ran Windows tools there under Wine, and a Ghidra run given that folder as its script path followed Wine's z: drive (the whole file system) into another disk, while the run's processes reached 60 GB of memory and the system's out-of-memory killer ended other programs. OpenCode can deny external_directory (*) with exceptions after it (the last matching rule wins), such as the original's folder in Deep mode; its shell's directories are found only best effort. Part of the approval decision above

ACP:

  • Gemini CLI, opencode, goose via ACP or native adapters; each verified live, then turn ACP on by default for the ones that pass
  • Over ACP, suppress session/update notifications until session/load answers, keeping replayed messages and tool activity out of the current turn — commit 38ffc39be084
  • Over ACP, say the turn's token count is unknown. The protocol's usage_update measures context, so it isn't added to per-turn or project token counts. — claimed: rudycelekli, branch bugfix/r2.2-acp-unknown-token-usage
  • Devin over ACP: devin acp asks the host to authenticate (meta.api_key) before session/new, so the backend's stored key is sent — claimed: NibrasAz7, branch feature/r2.2-devin-agent

Finding them:

  • #105, #135, #259, #385 and #423 predate 0.3.0, which looks in more places, and came through the ticket relay, so nobody can ask them for doctor output. Close them pointing at 0.3.0, and add search paths only for a new report that has that output
  • #423: an agent that is set up but switched off is never offered again, so "No coding agent was found" stays on screen with Claude installed: the setup card and "Look again" leave out every saved AI by name, on or off. Offer a switched-off agent as found and turn it on, keeping its options. Cause is not confirmed on the reporter's machine (needs their config.toml); the fix covers the switched-off case that the code leaves out (is_on, leave_out_set_up, builder_backend). — claimed: fabox-os, branch bugfix/423-disabled-agent-not-offered
  • Say plainly when what's found is the Claude desktop app rather than the claude program (#259, a guess at its cause)
  • macOS: check that Claude.app is where the desktop app is found (#259's hint is untried there)
  • #259's hint, the gaps left in review of #781: it's hidden when only ~/.ssh/config suggestions are found, a Store install without its claude-desktop alias isn't seen, and a desktop app's claude.exe on PATH is taken for Claude Code (#824) — claimed: FuryNocturn, branch bugfix/r2.2-claude-desktop-hint-gaps
  • macOS: /Applications/Antigravity.app ended up as the agent's command (#467). Found: the "Where is the program?" picker (pick_file) returns the bundle, and set_command saved any path. Refuse a folder or app bundle when it's saved (set_command), and don't run one that was saved before (program_of uses the usual name). Tests: an_app_bundle_is_refused_as_an_agents_program, program_of cases. — claimed: fabox-os, branch bugfix/467-agent-program-app-bundle
  • Windows: look in %LOCALAPPDATA%\agy\bin too (#62) — claimed: Afloat16, branch bugfix/r2.2-antigravity-search-20261103
  • Windows: usual_windows_dirs_come_after_path counts Antigravity's second folder (#748 added it; the test runs only on Windows, which CI doesn't)
  • Windows: setup and doctor take the user profile (USERPROFILE) for home, as the apps do, not HOME, which MSYS2 and Cygwin shells point at their own folder. — claimed: Plastic-time, branch bugfix/r2.2-windows-home-is-the-user-profile
  • CI runs the tests on Ubuntu only, so cfg(windows) tests (search paths, consoles, Job Objects) are checked only on contributors' machines: run them on a Windows runner too
  • "Claude Code on " is offered for any SSH host, whether or not it has Claude Code (#62)
  • Gemini CLI's sign-in reportedly refuses individuals now (#400): check, and stop recommending it in the builder setup dialog if so
  • DeepSeek Harness: an adapter runs dsh --profile headless "<task>" and hands it our tool server as a generated --patch overlay for that run (its own project, inbox and mode), since the harness reads tool servers only from profile layers; a question gets none and the person's own profile is left untouched. Verified with adapter and overlay tests and a live follow-up website build whose journal shows rebuild_check, core_unknown_add and core_handoff_write going through our server — claimed: 0231071, branch feature/dsh-agent-adapter
  • DeepSeek Harness: send tasks on stdin so multiline and long messages can start through the Windows desktop launcher (see the argv note above) — claimed: gtxy27, branch bugfix/r2.2-dsh-stdin
  • DeepSeek Harness: no resume (its headless profile keeps no session), and a build on the harness's desktop profile is untried
  • Apply repeated client selections and aliases once, preserving their first selection order. — claimed: Plastic-time, branch rrfz/bugfix-r2.2-deduplicate-clients
  • setup writes the plain mcp command, so it drops an --all-tools an entry already has, and doctor then calls that entry "a different copy or shape of the command" and suggests the very command that breaks it. The installed skill now warns against running setup on such an entry; doctor and setup should keep the person's extra arguments instead — claimed: 18923837179wjt-dot, branch fix/setup-skill-warns-before-setup-drops-all-tools
  • Windows: a .cmd that only runs one .exe by its full path ("C:\…\codex.exe" %*, written by hand or by another tool) was started through cmd.exe, which refuses an argument with a line break ("batch file arguments are invalid"), expands % and caps the line at 8191 characters; that .exe is now started directly, as an npm shim's is (#1237) — claimed: jgt87, branch bugfix/r2.2-cmd-wrapper-exe — PR #1238
  • Codex's and Grok Build's config.toml keep their Windows line endings when setup adds or removes our entry. — claimed: Plastic-time, branch bugfix/r2.2-toml-config-crlf
  • "Just ask" says before a question is sent when the project's builder can't answer questions (Codex, OpenCode and the other agents that can't be held to reading), and which AIs can, instead of refusing only after it's sent (#1695) — claimed: justinjohn0306, branch bugfix/r2.2-just-ask-says-first

Owner: maintainers · Status: in progress

  • UI tree on Linux (AT-SPI)
  • native_window_capture on macOS: no Screen Recording is refused before the program starts (it was "didn't show a window", which said to wait longer), and the name fallback matches whole words and never the system, so win no longer pictures the menu bar
  • native_window_capture keeps looking for up to 10 s past wait_secs when no window is there yet, instead of deciding from one look; tried with a program that opens its window late
  • Close a pictured program politely before killing it: a helper it started through launchd outlives the process group, and can stay running with the program's data files open
  • apple_calls_observe: the breakpoint callback is one valid line (it had a literal backslash and n between statements, so lldb stopped before launching, #251), covers every breakpoint instead of only the first, a failed lldb run is an error and not "0 calls", and names or arguments with line breaks are refused (they'd add commands to lldb's command file). The callback body was compiled with Python; lldb itself was not run
  • Run apple_calls_observe on a Mac against a small program with lldb (the #251 probe): the opt-in engine test records alpha and beta in order and reports a debugger refusal for a non-executable input. Verified with Apple LLDB 2100.0.16.4 on arm64 macOS 27.0.1 — PR #1844; claimed: rudycelekli, branch test/r2.4-verify-lldb-call-observation
  • Window capture tried on macOS and Linux (only Windows so far); games behind a launcher
  • Short scripted input sessions (enigo) to record screens of the original
  • A macOS app bundle picked as an app target (TipCalc.app) is studied as a plain folder: the study finds no program in it ("Couldn't find the program to decompile"), so Deep mode skips Ghidra, the console runs and the window picture. Study the program its Info.plist names — claimed: htcom-code, branch bugfix/r2.4-app-bundle-program
  • Since that fix, the console-run check reads an app bundle's Info.plist to name the original, in Clean room too (review on #1759). Clean room takes the picked path's own name — claimed: htcom-code, branch bugfix/r2.4-clean-room-bundle-name
  • A Windows game's folder picked on Linux or macOS (a 1990s Windows game: Game.exe, no executable bits) has no program to study: the folder search counts only programs this computer can start, so Deep mode skips Ghidra. Count a Windows .exe as a program to study on every system — claimed: youkaicountry, branch bugfix/r2.4-windows-program-in-folder

R2.5 Behaviour checks for non-web builds

Owner: open · Status: in progress

  • Keep mechanics behavior unverified when only the sheet has been read: no literal-use claims, persisted unknowns and an explicit usable-build summary; no builder retries for a missing oracle
  • Give console run comparisons a numeric-output oracle when needed: word likeness drops single-character tokens, so bare 4.5 and 9.0 both compare as empty output. An output that is only numbers is now compared as numbers, in order, when the original's repeated run can't say which ones vary: 4.5 against 9.0 fails the claim and says both, the same number still matches, and printing nothing where the original prints a number fails too. With words beside the numbers, likeness still decides, so a differing timing isn't a failed build — claimed: muraa-p, branch bugfix/r2.5-numbers-only-output
  • Compare console output's numeric spelling and order at positions stable across two original runs, and explain when varying numbers or unavailable repeats limit comparison — claimed: Afloat16, branch bugfix/r2.5-console-numeric-results-20261111
  • Behaviour checks that use the sheet: run the build with scripted input and measure values (speed, jump height, fire rate) against it (needs R2.4's scripted input)
  • Hold the self-test to what the check can see too: picture the game after its scripted input and require it to differ from the --screenshot picture (a report alone is the build's own word)

R2.7 The request decides how to build

Owner: open · Status: in progress

  • Recognize Phaser and Bevy requests in the stack catalogue so Make can start those games without a Unity, Unreal or Godot editor. — claimed: Afloat16, branch feature/r2.7-phaser-bevy-game-stacks
  • Make: a game whose request names a technology the stack table doesn't know (Phaser, Bevy…) still needs the default engine installed to start, though the builder is told to use what was named; ask the person, or let such a game start without an engine
  • The website check's missing-site note names every place site_dir accepts (dist/, build/, out/, or a plain index.html with no package.json) and the app's real place, "app/" or "the project folder" (#494)
  • A project edited in place (Work on my project, or rebuild --existing edit) keeps what its files are built with, unless the request names another stack: an engine's project file, then package.json and Cargo.toml dependencies, then found_by in stacks.toml, and a project whose dependencies name no stack there is left to the builder. A request that read as a game used to ask for Unity even in a Godot project, and an app got the kind's default — claimed: xerenon, branch bugfix/r2.7-stack-from-files; PR #755
  • A person's own project (edited in place, nothing to recreate) is judged by what it does for itself: its own build, then its own tests, through [[toolchain]] in engine/src/toolchains.toml (Rust, Node.js, Python, Go, .NET, Maven, Gradle, Make). Nothing is added to it for our sake: no --screenshot, --self-test or mechanics sheet. A site stack with nothing to recreate is a website, whatever the request says — claimed: xerenon, branch feature/r2.7-own-project-checks; PR #760
  • Run the Go, .NET, Maven and Gradle rows of [[toolchain]] on real projects, and every row on Windows (windows_program): Rust, Node.js, Python and Make were run for real, on a Mac only. Also a person's own Unity project: it's opened in batch mode, which should exit with an error when its scripts don't compile (not run: no Unity on that machine)
  • A person's own tests that already failed before their request fail the check too, so the builder is sent to fix them. If that proves a nuisance, run them once before the first turn and hold it only to what it broke
  • New projects build through the shared [[toolchain]] catalogue. Python root entrypoints, Go's root main.go, and one root .csproj also launch for screenshots, self-tests and Play; existing packaged programs, Rust, Electron and game-engine paths keep priority — PR #950 — claimed: Codex, branch bugfix/r2.7-new-project-toolchains
  • Runtime window capture follows runner descendants (including go run) and reads controls from the same process that supplied the picture, without falling back to unrelated names — PR #950
  • Shared toolchain commands make project-local executables absolute before changing the working directory, preserving virtualenv paths for relative project placements — PR #950
  • Launch a Go project with one cmd/<name>/main.go as a package when the root has no main.go — claimed: Afloat16, branch feature/r2.7-go-command-layouts
  • Launch the single installed console script of a Python project from its virtual environment — claimed: Afloat16, branch feature/r2.7-python-console-entrypoint
  • Extend launch recipes to Java build outputs and Python packages; build checks already use their toolchain commands. A single <name>/__main__.py runs as python3 -m <name>, the single runnable jar in target/ or build/libs/ with java -jar (Maven now builds with package); build wrappers stay out of launch — PR #1209; claimed: rmsff, branch feature/r2.7-python-java-launch
  • Audit relative project placements in build, option-run and Play paths too: runtime window capture already normalized its request paths; engine build arguments, option runs, console run checks, Play and pins now resolve the program, working directory and output file to absolute paths before starting a child in app/ — claimed: AlexJoaquimPereira, branch bugfix/r2.7-relative-project-paths
  • Decline a Godot launch when its project path cannot be represented as UTF-8 instead of silently replacing bytes in the selected path. — claimed: Afloat16, branch bugfix/r2.7-native-runner-project-args
  • Resolve Godot's project argument and the local Electron binary and project argument before starting either runner from a relative project folder — claimed: Afloat16, branch bugfix/r2.7-project-launch-paths
  • "player" alone doesn't make a request a game: a video or music player is an app (#953) — claimed: justinjohn0306, branch bugfix/953-player-not-a-game
  • System-tool and media-player phrases keep the app target: "a low level disk tool", "an MPEG 2 player" and "a single player app for movies". Independent game cues still apply — claimed: Afloat16, branch bugfix/r2.7-media-and-system-apps

R2.8 Clean room means clean room

Owner: maintainers · Status: in progress

  • A remote (SSH) builder needs a craft-anything-cli that knows --deep and --clean-room. Every release from 0.1.0 declares both as global flags and applies them in mcp, and has mcp --all-tools, the only flags a remote tool server gets; a test fails when mcp_for gives it another. Claude Code and Codex builds over SSH now check the host has craft-anything-cli before starting, as OpenCode's did — PR #1211; claimed: rmsff, branch test/r2.8-remote-cli-flags
  • A remote builder's craft-anything-cli should know CRAFT_ANYTHING_AGENT (#1022), which no release up to 0.5.0 reads: there, a CLI the builder starts calls tools as the person. Today that loosens nothing of the build's (the remote tool server has no project or inbox). Check it before a build once a release has it: version numbers can't tell (main says 0.4.0), so ask the CLI itself

R2.9 Everyday use: typing, projects, settings

Owner: maintainers · Status: in progress Gaps found while using the app for real projects.

  • Each project chooses, per kind of action, what the builder's approvals get (allow, ask or refuse): reading the web, commands, files outside the project, pushing and publishing, deleting outside the project, system changes, other tools. Unset kinds follow Auto mode as today. Found when Auto mode refused every git push without a card, so the person's OK in the prompt box never reached the check, and refused gh repo view as publishing too. Tried live by a maintainer on Windows 10 (a release build of main at b700e1d, 2026-11-06): a builder's push was approved and went through — claimed: tanish2k09, branch feature/r2.9-project-approvals

  • Agents that don't ask our approval tool (OpenCode --auto, Hermes -z, Maki's -p) aren't held to a project's approval choices either; part of R2.2's approval decision

  • While "commands" are allowed (Auto mode's default), a builder can run any program. Commands that name craft.json, the inbox or the approvals tool count as system changes, and our CLI started inside an agent calls tools as an agent, but a script that builds the path, or unsets CRAFT_ANYTHING_AGENT, isn't caught. Keep the person's choices where a builder's shell can't write them, or say plainly that "commands: allow" trusts the builder

  • When project approvals refuse an action, show the kind and how to ask before trying it again in the build feed. Repeat refusals of the same kind share one notice. — claimed: rudycelekli, branch bugfix/r2.9-approval-refusal-guidance

  • The project page and a slash command change those choices (like /btw): the Approvals menu under the prompt box and /approvals <kind> <allow|ask|refuse|default> — claimed: tanish2k09, branch feature/r2.9-approvals-ui

  • The new approvals strings in every catalog were written with an AI; not reviewed by native speakers yet — pt-BR claimed: marlonferreira1800-hue, branch docs/r2.9-approvals-pt-br-review

  • A prompt that changes nothing in app/ still builds and starts the app again, often twice (the builder's own rebuild_check, then the check after its turn). Keep a check whose round all worked while the app's files and what it's compared with are unchanged — claimed: tanish2k09, branch feature/r2.9-reuse-unchanged-check

  • A check starts the app in front, with focus, on the main monitor, in the way of whatever the person is doing. The --screenshot contract now asks for the window without focus, and on Windows behind other windows (egui's with_active(false) and AlwaysOnBottom, which winit shows with SW_SHOWNOACTIVATE at HWND_BOTTOM; on macOS and X11 a covered window is reported occluded and eframe stops painting it, so it isn't lowered there); it must keep drawing unfocused (Unity's runInBackground). Play still opens it in front. It's the builder's to do, so a build made before reads it only in a new brief, and nothing checks it was done. Not run live — claimed: tanish2k09, branch feature/r2.9-check-window-behind

  • A runtime study's check also starts the build normally (native_window_capture) to picture it beside the original, and that start still comes up in front with focus. Starting a program without focus needs STARTUPINFO's show mode, which std has only unstable (CommandExt::show_window, rust-lang/rust#127544), or a safe window crate to send it back

  • Keep New project's content column fixed while the vertical scrollbar animates in or out, hiding it when there is nothing to scroll, so switching tabs never shifts the page horizontally — claimed: Codex, branch bugfix/r2.9-home-scrollbar-gutter

  • Agent paragraphs with glossary terms and long file links wrap together without pushing their card or later messages under the sidebar (#523); Project and Ask bounds checked at two sizes — claimed: carl-dourado, branch hotfix/r2.9-523-message-width

  • Create atomic-write temporary files exclusively with short names distinct from the destination and private Unix permissions from creation. — claimed: Afloat16, branch bugfix/r2.9-atomic-temp-creation-20261103b

  • A prompt brought back with Up keeps its pictures (their paths are absolute, so they don't come back after the project is moved)

  • Change a project's builder from the CLI and MCP: project_builder_set, with default giving it back to the app's choice — PR #1833

  • Change a project's builder from its page, through that same tool (the app's own control)

  • chat --backend <a name that is turned off> and compare refuse it before model or agent setup: AiHub::start checks enabled and says how to turn it on or choose another — PR #1008; claimed: Codex, branch bugfix/r2.9-disabled-chat-backend (the API regression uses invalid tunnel settings to detect setup before the guard)

  • Ctrl+V (Shift+Insert, Cmd+V) in a prompt box that takes pictures pastes a copied image (saved as a PNG) or copied picture files; attachments show as square picture tiles, the whole picture on hover, decoded on a worker

  • A picture over 20 MB dropped on a prompt box or picked in its dialog is refused when it's added, as a paste does (workspace::attachable_picture), before anything is copied

  • When the builder decides a question, show the question and its choice as an answered card in the feed (the run sees only the tool call today; the builder's words say what it chose)

  • A question's own cost: a forked Claude Code session reports the whole session's running total, so a forked answer shows none; count questions in the project's usage — a question's cost is what its total added since the builder's last turn, unknown without a kept total; its cost is added to the project's usage, and concurrent usage writes in the app are serialized — claimed: AlexJoaquimPereira, branch bugfix/r2.9-question-cost

  • "Make this change" on an answer that suggests one (sends it as a normal message)

  • "Just ask" is told it has fewer tools than the builder and that the build may be running beside it, so it no longer reports the build's tools as disconnected or its work as cut off

  • "Change it" knows what was just talked about in "Just ask": the next message tells the builder the latest few questions and answers, once, and a Clean room build only those asked in Clean room (#857). Not checked in a live build

  • A note sent while a build is running tells it what was asked in "Just ask" (#857): the note's hand-over file carries the answers, and the after-turn path puts them before the notes; only answers the run's mode may show, marked told when read or sent — claimed: AlexJoaquimPereira, branch bugfix/r2.9-note-just-ask; commit 78d3963

  • A note still waiting when the app closes is sent by the next run instead of deleted when it starts (#1620): its hand-over file keeps the person's words (notes::Left), and a run started without a message takes them as its own waiting notes, so they stay on disk until its builder takes them; a folder it can't read is moved aside (notes-unread-…), not deleted. A file from before this has only the hand-over text, whose Just-ask answers may be from the other mode, so it's still dropped. Unit-tested; not run live — claimed: justinjohn0306, branches bugfix/r2.9-notes-survive-restart, bugfix/r2.9-left-notes-kept

  • Record which mode a builder's session ran in: agent_session is kept across modes, so a Clean room build or question can resume a session from a Deep build and carry what it read, and such a question's answer is recorded as asked in Clean room. Don't resume one for Clean room — agent_session_deep in craft.json records it (one saved before the field counts as Deep); a mismatch starts from the brief, and a question is answered from the files; tests cover both directions and the legacy case. Claimed: AlexJoaquimPereira, branch bugfix/r2.9-session-mode

  • The shared investigation chat (SharedConversation) doesn't continue a Deep chat's agent session once the chat is in Clean room; Clean room to Deep still continues — claimed: rmsff, branch fix/r2.9-shared-chat-mode

  • A session that opens a second project kept the first project's evidence and open questions: open_project added the new project's records to the session ledger without clearing the old, so core_evidence_show, a report and the evidence ledger mixed two projects. Opening a different project now clears the ledger, order, open questions and unloaded count, then loads the new project's store; reopening the same project still refreshes it in place. The regression in session.rs covers the switch, the same-project refresh and returning to the first project — claimed: AlexJoaquimPereira, branch bugfix/r2.9-project-evidence-switch

  • A project opened after another keeps the session's open targets and their annotations: a report for the new project can name the old project's target, and annotations saved after the switch go to the new project's store. Decide whether opening a project closes the targets or reloads their annotations from the new one

  • A Clean room build or question must also not resume an API backend's kept transcript: ws.transcript_path() is loaded whenever it exists, so switching modes can carry what a Deep API build read. Clear it, or record its mode, when the mode changes; add a regression — an API transcript records the mode it ran in (TRANSCRIPT_VERSION 2); one from the other mode, or a version-1 one (counts as Deep), isn't resumed. A same-mode continuation keeps its transcript and carries on; a mismatch studies again from the brief. Claimed: AlexJoaquimPereira, branch bugfix/r2.9-api-transcript-mode

  • Study can still be skipped after a mode switch: a coding agent's builder_started() was mode-unaware, and an API follow-up message with an existing capture left first false even when the transcript was rejected. Gate both on the run's mode — builder_session_in matches the saved session's mode, and a run studies again when its only kept conversation is in the other mode; a follow-up with no kept conversation keeps its old path — claimed: AlexJoaquimPereira, branch bugfix/r2.9-mode-switch-study

  • Exercise a mode-switched API follow-up end to end: an other-mode transcript, a capture and a message must study again, not only the decision helper

  • Record which mode wrote capture/, so a failed run that leaves a capture from the other mode isn't handed to a same-mode conversation; remove Deep-only capture files when a Clean room study replaces them — claimed: AlexJoaquimPereira, branch bugfix/r2.9-capture-mode

  • A saved Claude --resume/--continue in the backend's extra arguments is refused in a Clean room build or question when it would reuse the project's Deep session, leaving unrelated arguments alone; unit and scripted CLI tests cover the build and question paths (Clean room refuses, Deep allows), but no live signed-in Claude run is done, so this stays open — claimed: AlexJoaquimPereira, branch bugfix/r2.9-saved-resume

  • #504: the builder's plan card can be folded away with its X, or with project_plan_fold (between builds: a fold during a build can be overwritten by its next save). The steps stay in the conversation (FEED_VERSION 4, FeedItem::PlanDismissed), and a later run's checklist shows again — claimed: claude, branch bugfix/r2.9-plan-card-dismiss

  • A run that ends with steps still open marks them as stopped on its card, instead of showing "Doing" until someone folds it (#504: Claude's runs left the plan unticked) — claimed: rust-craft, branch feature/R2.9-stop-open-steps; PR #876

  • #666: a long line in a feed's tool-work row (a Deep-mode decompile, a status note) ran off the window and pushed the feed left; the header title can't wrap in egui, so it's the note's first line now, cut at 90 characters, and the full note wraps in the body — claimed: claude, branch bugfix/r2.9-feed-text-overflow

  • decision A program started from the CLI (project_play) keeps the shell's output pipe open, so the command doesn't return until it closes (#207). Every fix found needs unsafe or an unstable API: an exception to rule 9, or a limit we document. Meanwhile a reproducing test marked #[ignore] and a line in the CLI docs are welcome

  • macOS: the window drew Windows-style controls at the top right, since the app turned off the system's title bar (with_decorations(false)); it now keeps the system's buttons, frame and resize edges under a see-through title bar, which is 32 pt there so the buttons sit centred in it (#20)

  • macOS: the app's icon is bigger than the system's (#20)

  • Show a command's output while it runs: Claude Code reports a tool's result only when it ends, so a long cargo build is one quiet line until then. Watching the agent's child processes from outside could say what's running and for how long

  • Release the keep-awake helper before a rebuild reports Finished, waiting up to five seconds for its supervised cleanup and logging a timeout. Real subprocess stand-ins checked on Linux (#541) — claimed: Afloat16, branch bugfix/r2.9-keep-awake-cleanup-20261103

  • Recheck keep-awake cleanup with a real Windows CLI rebuild: the helper once outlived it by more than 9 minutes and held the caller's output pipe; not reproduced in 8 more runs. Found: closing the desktop app during a build ends it without dropping the guard, and the helper (an endless PowerShell loop, started with a week's time limit) carried on. It now ends by itself once the process it serves is gone: the Windows loop checks that process every 30 s, and macOS uses caffeinate -w. Live on Windows: a CLI build killed outright, its helper ended 30 s later. Not run on a Mac

  • Zoom (Cmd/Ctrl + / − / 0) between 50% and 200%: egui's own limits are 20% and 500%, where text is unreadable or the window has room for almost nothing — claimed: sanketdofe, branch feature/r2.9-zoom-bounds

  • Find why a site capture lost its browser ("the WebSocket is closed", 419 ms into web_site_capture for a second page, seen once; the same capture passed 41 s earlier)

  • Delete a private browser's temporary profile when it closes or fails to start (#810) — PR #822

  • Stop private browsers (and Electron apps we drive) that outlive the app after an abrupt exit, and delete their profile (#810): a watcher stops the browser's process group once the app is gone (platform::end_with_this_process, Unix). Unix unit tests added; its script was run against real process groups, but not on a Mac with Brave — claimed: justinjohn0306, branch bugfix/r2.9-browser-ends-with-app (stopping private browsers)

  • Investigate the Brave Helper crash on launch (#810); needs macOS reproduction

  • Check whether a private browser outlives an abrupt exit on Windows (its Job Object may not be killed when closed); add a watcher there if it does

  • Claude Code's Ultracode switch (a setting, not an effort level; the effort environment variable we use can't turn it on)

  • A model the list doesn't offer (opusplan, [1m] variants, one another harness serves) can be typed in beside the model list in Settings → AI, as ai model already allowed (#470)

  • Claude Code's model follows the person's own setup: the backend no longer passes a hard-coded --model (Opus 5.5) on every run, so a model customised in Claude Code's own settings applies. Settings → AI offers "Default (its own settings)" first in the list, then the built-in ids; a custom id can still be set with ai model <name> <id>, and ai model <name> default or --default clears the choice the same way. The args, the CLI and the settings picker are unit-tested; not run with a live Claude Code build

  • The preview server answers only requests addressed to this computer (a Host and Origin check against DNS rebinding), streams large files instead of reading them into memory, sends no body for HEAD on a missing file, and caps open connections and the time a request may take to arrive (#597, #598, #600, #601, by Afloat16)

  • The command palette stays where the wheel scrolls it; it follows the selection only when the arrow keys or typing move it (#584)

  • Linux: draw on the GPU that drives the screen. egui-wgpu's default picks the discrete GPU, so on a laptop whose integrated GPU drives the panel the window took 20 s to open and the whole desktop lagged (#571). Pick by the connected displays in /sys/class/drm: the built-in panel's GPU, else the only GPU with displays, else the discrete one; WGPU_POWER_PREF and WGPU_ADAPTER_NAME still decide when set. Checked live on an AMD + NVIDIA laptop only (#578) — claimed: Nghiathan13, branch bugfix/r2.9-linux-gpu-choice

  • Linux GPU choice on a desktop with monitors on two GPUs, and on an Intel + NVIDIA laptop: covered by unit tests on fake /sys/class/drm folders, not run on such a machine yet (#571)

  • "Work on my project" on New project, beside "Make my own version" and "Just look inside": pick the folder of a project you already have, say what to change, and its files are edited in place, with nothing asked; the project takes the folder's name, an empty folder is refused and a folder that's already a project opens. Until now that was only reachable through "Choose a folder…" under Make's Start, on a screen that asks what to recreate — claimed: xerenon, branch feature/r2.9-work-on-my-project; PR #686

  • Leaving "Work on my project" forgets a folder it would edit in place and drops a folder dialog that ends later, so a version made on Make can't edit that project, and an address in its request isn't read as an original to recreate (review of #686) — claimed: xerenon, branch bugfix/r2.9-own-project-review; PR #750

  • build_files_list lists an edited-in-place project from the project folder itself, not a missing or nested app/, and leaves Craft Anything's own records out — claimed: AlexJoaquimPereira, branch bugfix/r2.9-build-files-in-place; commit 1ffeb40

  • build_file_read read the whole project file into memory before paging, so a large generated file under app/ could freeze or kill the app; it now refuses a file over 16 MiB with a plain error, while files at or under the cap page as before. The regression covers the cap and the boundary — claimed: AlexJoaquimPereira, branch bugfix/r2.9-build-file-read-cap

  • build_file_edit also reads the whole file into memory; cap it the same way

  • Choose Claude Code's folder (CLAUDE_CONFIG_DIR) per AI, so one computer can build with two Claude accounts; an app opened from the Dock never sees the shell's variable. Settings → AI (a typed folder, and Add another) and ai folder; live-checked from the CLI with an empty folder (not signed in, and the fix names the folder) overriding the shell's own variable, and by a build in the app with a typed ~/ folder. A sign-in that expired says so instead of "isn't signed in" (the folder's .claude.json still names the account) — claimed: Claude (Opus 5.5), branch feature/claude-config-dir

  • A live build with two Claude Code AIs signed in to two accounts, and one over SSH with a ~/ folder (the remote line is unit-tested, not run on another computer)

  • The Advanced console's status stays clear of its tabs when the console is narrow — claimed: justinjohn0306, branch bugfix/r2.9-console-status-overlap

  • Keep the builder-availability smoke regression independent of live agent discovery. — claimed: Afloat16, branch test/r2.9-builder-discovery-fixture-20261104 Test fixture implemented with controlled discovery; automated validation remains pending.

  • macOS: a menu bar with the app's own actions (open a file or folder, new project, settings, Simple / Advanced, Clean room / Deep, the command palette, help); the system's menu bar showed only the app menu — claimed: sanketdofe, branch feature/r2.9-macos-menu-bar

  • A Traditional Chinese README (README.zh-hant.md) translated from README.md and linked from the language switcher in every README — claimed: monarchsofchaos004-gif, branch docs/readme-zh-hant; PR #926

  • macOS: an Edit menu (Undo, Cut, Copy, Paste, Select All). The keys work in text boxes, but the system's Edit items need the text box to answer them, so they'd show greyed out today

  • The Advanced console's prompt box starts tall enough to read what's typed (#942) — claimed: justinjohn0306, branch bugfix/942-console-input-height

  • A French README (README.fr.md) translated from README.md and linked from the language switcher in every README (#831)

  • A Thai README (README.th.md) translated from README.md and linked from the language switcher in every README

  • Vietnamese letters with stacked marks (ạ ả ấ ế ơ ư ữ…) aren't in the bundled Figtree or egui's default fonts: without craft-fonts they show as boxes, and with it each such letter comes from Noto Sans CJK while the rest of the word is Figtree. Bundle a build of the text font that has the Vietnamese set (with its OFL notice) so Tiếng Việt renders in one face

  • A build told craft-fonts are required (CRAFT_FONTS_REQUIRED=1) finished with none embedded when CRAFT_FONTS_DIR was missing: the build script read the switch only when the folder was set and bad. It fails now for no folder, a bad one, and a manifest that lists no fonts; CRAFT_FONTS_REQUIRED set to nothing or 0 means not required. The decision is plan in crates/ui-egui/build.rs, tested from tests/craft_fonts_build.rs. Built for real on Windows only — claimed: hichang4u, branch bugfix/r2.9-fonts-required

  • A required browser font build fails when a nonempty craft-fonts manifest has no fonts left after the WASM filter (#1736 review). Empty-manifest and missing-file errors stay distinct; native and optional builds keep their behavior. Build-plan tests and real wasm32 Cargo fixtures checked on Windows; no browser rendering checked — claimed: Butler12345, branch bugfix/r2.9-wasm-required-font-filter

  • An engine download the server can't answer right now (a 5xx, 429) says to try again later, not "update the pinned URL", which nobody installing an engine can do (#1193) — claimed: justinjohn0306, branch bugfix/r2.9-download-server-errors

  • Keep shared engine downloads on HTTPS across every redirect, with a real TLS regression and no signed redirect links in refusal diagnostics. Six controlled HTTP/TLS tests check rejection before an HTTP request, successful HTTPS redirects, hashes and file protection; the downgrade test fails without the HTTPS-only policy. Checked on Windows — claimed: Butler12345, branch bugfix/r2.9-https-engine-downloads

  • On Windows, engine downloads go through the person's manual system proxy when no proxy variable is set; they only read HTTPS_PROXY and the like, so behind a VPN that sets the system proxy the Godot and Ghidra downloads timed out (#1787, #1827). Only engine downloads; AI clients are unchanged. The setting's forms are unit-tested; not run behind a real system proxy — claimed: justinjohn0306, branch bugfix/r2.9-windows-proxy-downloads

  • Settings → AI: a card for an AI found on this computer wraps its description beside its Add button; the line didn't wrap and ran on under the button — claimed: jgt87, branch bugfix/r2.9-detected-cards-wrap — PR #1235

  • Settings → AI: Providers, Harnesses and Detect share consistent spacing below the discovery heading and wrap together, so the Detect action cannot overlap a tab — claimed: istaari, branch bugfix/ai-detection-spacing — PR #1530

  • The same action reads the same everywhere: a pin's menu said "Show folder" for what the project page calls "Open folder" (both project_open_folder), and a game's chosen folder was shown as a full path in its row, where other chosen folders are shortened, cut to fit and whole on hover — claimed: jgt87, branch bugfix/r2.9-same-words-open-folder — PR #1243

  • Colours, corners and text sizes off the theme in the chat, evidence, prompt box, sidebar, palette and Just ask: the Send button's own white and black, two hand-set corner radii, code at 12.5 pt where the rest is 12, an evidence summary at 13/15, and a question drawn in the mode's default size where other card titles are 15 semibold — claimed: jgt87, branch bugfix/r2.9-theme-tokens-type-scale — PR #1279

  • A build's AI cost estimate counts each Claude Code turn's own cost: a resumed session reports its running total, so the build summed the whole session at every turn (#1187). A turn's cost is now its total less the total kept for its session; a resumed session with nothing kept has no cost. Live check (2026-11-09, Haiku 5.5): the second turn's own tokens cost $0.00146751, and the reported running total was $0.00350053. Earlier builds' totals stay as they were. — claimed: fabox-os, branch bugfix/1187-claude-resumed-cost

  • A Deep-mode build told the builder "You haven't decompiled anything yet" after every request from the person, though it had read decompiled functions in an earlier run: the flag lived only for one run, and each request starts a new one. It is now kept in craft.json (decompiled), so a run reminds the builder only while it has never decompiled in this project. Unit test only; not run live — claimed: NeuTroNBZh, branch bugfix/r2.9-decompile-reminder-every-request

  • #1421: a run after the first skipped the study, so in Deep mode its builder's tool server had nothing open and the first native_* calls said "Nothing is open yet". Such a run now opens the program the study looked at (core_target_open, only when nothing is open, never in Clean room); a failure is said in the feed and the builder can still open it. Unit test only; not run live — claimed: NeuTroNBZh, branch bugfix/1421-reopen-project-target

  • rebuild_check said "no build is running" when the run had taken the check and one of its steps hadn't finished: on a Windows Unity build, three checks in a row stopped at "game-self-test" and the builder gave up after its default 20 minutes, below the run's own 30-minute limit on a check. The run now marks a request it takes, the tool tells "still going" (TimedOut) from "no run took it", and its default wait (35 min) outlasts the run's limit, so the run reports a stuck check itself. Unit tests only; the stuck self-test wasn't reproduced (the same game's --self-test now exits in 16 s) — claimed: NeuTroNBZh, branch bugfix/r2.9-rebuild-check-wait

  • Review of #1417: an answer landing as rebuild_check's wait ended could still read as "no build is running", since the run removed its mark when it answered; only the asker removes it now. A build stopped during a check answers it ("stopped before this check finished") instead of leaving the builder waiting, including a stop seen before the check starts — claimed: NeuTroNBZh, branch bugfix/r2.9-rebuild-check-mark-race

  • Find why a builder's own check stayed on "game-self-test" for over 20 minutes on Windows, past the run's 90 s limit on the game's run (seen three times with a Unity 6000.6 build)

  • Let waiting notes scroll with the conversation so a long note leaves the live build activity and prompt box reachable — claimed: jeffreylm, branch bugfix/r2.9-keep-build-activity-visible

  • Re-running at a folder that already holds a project keeps its saved builder, asset budget, generate_assets and stack, and uses the typed request as a follow-up instead of dropping it (regression from #987); new-project-only options are refused on an attached project — claimed: AlexJoaquimPereira, branch bugfix/r2.9-reattach-keeps-settings

  • A failed external program's error says "printed last" from the first stderr_cap bytes and never says the rest was dropped; keep its last output for the message and say when it was cut — claimed: AlexJoaquimPereira, branch bugfix/r2.9-stderr-tail

  • After #1682: an attached project's run uses its saved builder rather than the Make draft's, keeps its live feed when it is already building, and keeps a queued request with its options; a start that fails keeps the Make draft — claimed: AlexJoaquimPereira, branch bugfix/r2.9-attached-builder-feed

  • A queued run's request lives in its RunOptions and the in-memory queue; an app exit before it starts loses it. Let the run side keep it (a queue record or a left note) across a restart

  • A config.toml that is there but can't be read (another program holds it, no permission, not UTF-8) loaded as defaults like a missing one, and the next save replaced it; one that couldn't be parsed was renamed aside with the rename's result ignored. Only "not found" is plain defaults now. A file that can't be understood (not TOML, not UTF-8) is set aside whole under the first free name (config.toml.broken, .broken.1… up to 16). One that can't be read, or can't be set aside, stays where it is, and the defaults standing in for it (Settings::not_loaded) refuse save_to and a deferred save, also after settings_set's round trip. Tested on Windows with a file held without sharing; the Unix permission test wasn't run here — claimed: hichang4u, branch bugfix/r2.9-config-not-loaded

  • When config.toml wasn't loaded (Settings::not_loaded), the app says so at startup where the person is looking, with the file's name and why, and that nothing is saved over it (settings_ui::not_loaded_warning); English only for now, like the builder notice

  • A config.toml that can't be read for a moment at startup (a virus scan or another program holding it) isn't tried again, so the whole session runs on defaults and refuses every save: try the read again for a short while before giving up — claimed: NicolasBohn, branch bugfix/r2.9-config-read-retry

  • A website build has no dev or start script, since Craft Anything serves its built folder itself, so outside the app there was nothing to run (#1590): the builder writes app/README.md saying how to build it and which folder to serve — claimed: justinjohn0306, branch bugfix/r2.9-site-readme-how-to-run

  • An OpenAI-compatible, OpenAI, OpenRouter or Gemini AI has no effort setting, and nothing asks it to think harder (#1813): Settings → AI and ai effort offer its reasoning levels, sent only once the person picks one — claimed: justinjohn0306, branch bugfix/r2.9-api-reasoning-effort

  • A website study can see what the person sees when signed in: the person signs in once in a visible window (web_site_sign_in), the saved sign-in is reused by later opens of that exact site (copied into the private profile, never changed by a capture), and web_site_sign_out forgets it; the builder is told when the capture was signed in. Addresses #1178. Live with Chromium on Linux against a local test site (virtual display, scripted sign-in); a confined Snap or Flatpak browser and a real site were not run. The window's own chrome://version and chrome://prefs-internals showed the Default profile and password and card saving off (Chrome for Testing 156, live test the_sign_in_window_runs_the_prepared_profile_with_saving_off). The live sign-in tests also passed twice with Chrome for Testing 157 on macOS, including the profile and session-cookie handoff; the Linux-only no-desktop test returns early there. Windows was not run — claimed: execsumo, branch feature/r2.9-signed-in-website-study Review fixes: saved profiles use normalized host and effective port; browser reuse also checks the saved revision, and explicit signed_in: false always starts fresh. Builder guidance treats saved-profile use as unverified authentication, even without a password page. The interactive sign-in runs without holding the engine session lock; same-site profile operations fail fast while reserved, without blocking other sites or explicitly fresh opens.

  • A "Sign in to this site" button in the Make screen for website targets (web_site_sign_in is reachable only as a tool, from the CLI, MCP or the agent, today)

  • Study with the person's own already-running browser through its debugging endpoint, as an alternative to a saved sign-in

  • Sign in once with web_site_sign_in on a real site behind a login (and with the Ubuntu Snap Chromium, whose launcher couldn't start in the test shell) and rebuild it signed in

  • Refresh a saved sign-in from a capture session when the site renews it, so it lasts longer than the first sign-in's cookies

  • Before the builder starts on a signed-in study, show the person the study's first screenshots and ask them to confirm it's the signed-in app (Make screen and an inbox approval)

  • A Rust build starts on a computer without Cargo and spends a whole builder turn before its build check says cargo is missing (#1837): a new build whose stack names a program (program in stacks.toml) stops before any turn when it isn't found, saying how to get it or to start again with another stack — claimed: justinjohn0306, branch bugfix/r2.9-check-cargo-before-build

  • The check for a stack's program was skipped for every SSH builder, but an API builder over SSH (a tunnelled model) builds on this computer (review of #1846): skip it only when the builder's tools run on the other computer (tools_run_remotely) — claimed: justinjohn0306, branch bugfix/r2.9-cargo-check-api-over-ssh

  • A project moved to another stack mid-build keeps its recorded stack (only package.json updates it), and the build check runs cargo build whenever app/Cargo.toml is there, so a Rust project rewritten in Python is still checked with Cargo until the old manifest is moved out (#1837): record the stack from what was built for every toolchain, and check with that

  • One design system for the desktop app: a typed scale (text sizes, gaps, padding, icon and control sizes) in theme.rs, one set of buttons (primary, secondary, ghost, danger), switches and fields in widgets.rs, and every screen moved onto them, so sizes and spacing stop differing from screen to screen; a test keeps literal sizes, paddings and colours out of the screens (docs/design-system.md). With it: Figtree for text; corners of 4, 6 and 8; fields and buttons a shade lighter than their card, with one soft edge; dropdowns, menus and the project's side panel as lists of rows; messages in the title bar; a sliding segmented switch; sharper small text and a wheel step of 100 points. Checked on every snapshot scene in both themes and in the app on Windows; not on macOS or Linux — claimed: bflatastic, branch refactor/ui-design-system

  • Design system, left for later: links come in two sizes (widgets::link follows the text around it); the title bar keeps its own control height (30, or 26 on macOS) instead of a ControlSize; frames with four different sides (prompt box, sidebar, workbench rails) still write their padding where it is used; the macOS disk image's background is still drawn in DM Sans and JetBrains Mono; dropdowns have no name for screen readers, only their value; fields and dropdowns have no 3:1 border at rest (WCAG 1.4.11), by choice, which wants a second look; a compact field is about a point taller than a button

R2.10 Pinned builds

Owner: maintainers · Status: in progress Keep a build that works at hand: pin it, and open it from the sidebar whatever the builder does to the project next.

  • Say when a light pin ended up copying what it meant to share (a drive without hard links): copied_shared_bytes on Pin, counted when a Light pin's file couldn't be linked, and said in a toast with the size (a_shared_file_that_cannot_be_linked_is_counted_as_copied) — in PR #220
  • Pins of one project sharing with each other, so a third frozen copy of the same assets takes no more space
  • Pin a website (a frozen copy of the built site, each served on its own port)
  • A pin started before the app restarted shows as running (keep its process id and start time in the pin, and check both), and can be stopped
  • Cap a run's log (last-run.log, play.log): it grows as long as the program prints. The program keeps writing straight to an append-only file, and while we are alive a watcher cuts that file back to empty once it passes 1 MiB, so its writes never fail or block and a program that outlives us is untouched; past our exit the file grows again, as it did before. — claimed: muraa-p, branch bugfix/r2.10-cap-run-log
  • The log-cap test assumed a cut lands within a few seconds of the stand-in's printing, so a loaded machine could grow the file past the bound with nothing wrong. The stand-in now reads its own log and stops once it has seen a cut and written past it — claimed: muraa-p, branch test/r2.10-log-cap-waits-for-cut
  • A slow pin size estimate that finished after the person cancelled or opened another project's dialog was applied to the open dialog: the estimate and its errors now name their project, and the dialog ignores a result for another project or a closed dialog. The result and error detail carry the project; UI regressions cover a late estimate, a late error and a closed dialog — claimed: AlexJoaquimPereira, branch bugfix/r2.10-pin-estimate-project
  • Opening a pin that's already running brings its window to the front, instead of a message
  • How much space pins take in Settings — claimed: Afloat16, branch feature/r2.10-pinned-copy-storage

M3 Engine depth

An open lane: anyone may take these at any time.

R3.3 Frida sessions for real

Owner: open · Status: todo

  • Replace the in-memory session bookkeeping with real attach, resume and detach
  • The Frida and adb output parsers cut a 64 KiB line at a byte index, so a multibyte character at the boundary keeps the whole line; cut at a character boundary and test both (#1095 follow-up) — a shared capped_line helper walks back to a character boundary; regression tests cover both parsers and the helper — claimed: AlexJoaquimPereira, branch bugfix/r3.3-engine-line-cap
  • adb parse_ps reads its header line before the 64 KiB cap, so a hostile header can materialize a full token list; cap it like the body lines — claimed: Afloat16, branch bugfix/r3.3-cap-adb-process-header

R3.9 Game assets out

Owner: open · Status: in progress

  • Unity: sounds out (the stored stream, usually FSB5; PCM ones as WAV) — claimed: NicolasBohn, branch feature/r3.9-unity-sounds

  • Unity: crunched textures (DXT1Crunched, DXT5Crunched; two of the games checked use them)

  • Unity: decode R16 and RG16 texture pixels through the shared bounded PNG export path. — claimed: Afloat16, branch feature/r3.9-unity-integer-textures-20261109 Source 540b4947aa1b; format checks and compatibility gates passed.

  • Unity: the float formats (RHalf, RGHalf, RGBAHalf, RFloat, RGFloat, RGBAFloat) — claimed: Afloat16, branch feature/r3.9-unity-float-textures

  • Decode Unity RGB9e5 shared-exponent textures through the existing bounded picture exporter. — claimed: Afloat16, branch feature/r3.9-rgb9e5-textures-20261110

  • Unreal 5 and Oodle: IoStore packages, unversioned properties and Oodle-compressed paks

  • Game files: save pictures from IoStore containers (needs the Unreal 5 support above)

  • Decode Unreal packed 10-bit normalized and 11-bit floating-point textures through the bounded picture exporter. — claimed: Afloat16, branch feature/r3.9-unreal-packed-textures-20261111

  • Unreal: decode cooked half- and full-precision floating-point textures with the shared bounded pixel reader. — claimed: Afloat16, branch feature/r3.9-unreal-float-textures-20261109 Source 75b507c0b96f; format checks and compatibility gates passed.

  • Unreal: sounds and meshes out (so a Deep-mode rebuild needs no outside converter for them)

  • Decode Unity EAC single-channel and dual-channel textures through the bounded picture exporter. — claimed: Afloat16, branch feature/r3.9-unity-eac-textures-20261111

  • Decode Unity ETC1 RGB textures through the bounded picture exporter, including both block modes and clipped edges. — claimed: Afloat16, branch feature/r3.9-etc1-textures-20261110

  • Decode Unreal DXT3 (BC2) texture blocks, including explicit alpha and clipped edge blocks. — claimed: Afloat16, branch feature/r3.9-dxt3-textures-20261110

  • Decode Unity RG32, RGB48 and RGBA64 textures without discarding their 16-bit channel layout. — claimed: Afloat16, branch feature/r3.9-unorm16-textures-20261110

R3.11 Single-file apps, game files and stand-in backends

Owner: open · Status: in progress

  • Bound native value-flow decoding and incoming parameter nodes by the requested graph node budget, and mark omitted code as partial. — claimed: Afloat16, branch bugfix/r3.11-value-flow-budget-20261110
  • Resolve AArch64 address evidence only from known operands, with the right arithmetic width and retained writeback effects. — claimed: Afloat16, branch bugfix/r3.11-a64-address-proofs-20261110
  • AArch64 shifted and extended register operands lose their shift: sub w9, w9, w9, lsl #3 lifts to w9 -= w9; and add w9, w10, w9, lsr #31 to w9 = w10 + w9;. Keep the shift or extension on the operand and show it in pseudo-C — claimed: htcom-code, branch bugfix/r3.11-a64-shifted-operands
  • AArch64 memory operands with an extended index (ldr w0, [x1, w2, sxtw #2]) drop the extension: the index reads as the plain register, so a sign-extended index looks unsigned in pseudo-C — claimed: Jensen-Yao, branch bugfix/r3.11-a64-memory-index-extensions
  • Preserve AArch64 exclusive-memory status results and paired load definitions without presenting them as unconditional stores. — claimed: Afloat16, branch bugfix/r3.11-a64-exclusive-value-flow-20261110
  • Keep the first 100000 byte-search matches in file-offset order across hex, UTF-8 and UTF-16LE queries, including capped pages. — claimed: Afloat16, branch bugfix/r3.11-byte-search-match-order-20261110
  • Validate the remaining loopback fixture corrections for complete and fragmented request headers, negative probes and stand-in cancellation on the current main — claimed: Afloat16, branch test/loopback-fixture-stability-20261103 Revision cd468aca, synchronized with main 41d2175e, passed full Linux CI. The fragmented-header regression withholds the final byte until the next read and keeps the existing request and response assertions. The request-encoding, negative-probe, cancellation and builder-discovery smoke checks passed in the same run.
  • Verify stand-in session cleanup by the original answerer's lifetime rather than polling a released port that another listener may reuse — claimed: Afloat16, branch test/r3.11-standin-session-lifetime-20261103
  • Stand-in backend: writes that change later reads (recorded answers are replayed as they are)
  • Tell the builder about the stand-in backend in a website's brief, when the capture kept a recording (capture/*.har): what web_standin_build, web_standin_start and the rest do (brief::standin_words); Deep mode only, since Clean room reads nothing inside the original
  • A Deep website study records the start page's traffic and the answers it gave into capture/original.har, so the brief's stand-in offer has a recording to point at (analysis::web_har::har_from_network, browser_tools::write_recording)
  • Live-check a Deep website study's recording on a real site, and the stand-in built from it
  • Start a stand-in backend from a recording automatically when a build begins
  • Stand-in backend: retain 304 samples without a usable cached body for diagnostics, skip them during replay, and report cached_body_unavailable when the best matching group has no usable answer. Build results count them; cached bodies still answer as 200 (#543) — claimed: Afloat16, branch bugfix/r3.11-standin-empty-304-20261103
  • Godot format 4 and IL2CPP metadata 38 and 39 are header-only; read their tables once a real file can be checked
  • Bun: take each module's stored source map and bytecode out as files. — claimed: Afloat16, branch feature/r3.11-bun-module-payloads
  • A standalone .pyz: a PyInstaller PYZ archive given on its own is its own packed reader (pyinstaller_pyz), listing the program's modules and handing one back as a .pyc; a Python zip application named .pyz stays with the ZIP reader — claimed: muraa-p, branch feature/r3.11-standalone-pyz
  • Collect native function evidence from every body block and infer parameters from its actual entry block. — claimed: Afloat16, branch bugfix/r3.11-native-function-block-evidence-20261110
  • native_source_folder_write uses the built-in decompiler only: use Ghidra in Deep mode when it's there
  • Keep source registers alive when an AArch64 arithmetic result is discarded into the zero register. — claimed: Afloat16, branch bugfix/r3.11-a64-discarded-results-20261111
  • Keep stand-in response samples distinct when headers, content types or body-availability metadata differ. Compare header names without case and ignore ordering between different names, preserving each name's value order. At the sample cap, a usable answer can replace an unavailable 304 for the same path, query and request key; other requests keep their missing-body evidence. Refused bodies cannot replace that evidence. Compare complete samples before charging retained bytes, using #717's retained-budget accounting. Historical revision f79dc0bd passed full Linux CI: all five original metadata regressions and nine header and capacity regressions passed. All 24 combined replay regressions pass locally on the refresh onto cf414ff4. See this PR's checks for the refreshed revision's complete gate. — claimed: Afloat16, branch bugfix/r3.11-standin-sample-metadata-20261103b
  • Keep evidence-based call paths within one input binary, rejecting mixed or missing source content identities. — claimed: Afloat16, branch bugfix/r3.11-call-path-input-identity-20261110
  • native_publisher_show hashes the file minus the signature fields, not section by section; a file with unusual gaps between sections can read as different — the hash now follows Windows' order (headers, each section by its data's file offset, then the bytes after them, which Windows counts without the certificate table), the tool no longer warns that a gap could read as different, and a regression puts a gap between sections — claimed: muraa-p, branch bugfix/r3.11-publisher-hash-sections
  • Check the section-by-section file hash against a second implementation: the test kit signs fixtures with its own copy of the steps (gaps, section order, data after the last section, sections that share bytes), and a file that declares more sections than were read isn't hashed — claimed: scks1, branch bugfix/r3.11-authenticode-section-hash Real signed files: 69 still match on Linux (STATUS); none has gaps between sections
  • Preserve source registers and link-register writes for AArch64 returns and authenticated control transfers. — claimed: Afloat16, branch bugfix/r3.11-a64-control-registers-20261110
  • AArch64 move-wide immediates lose their value: movk w9, #0x9249, lsl #16 lifts to w9 = 0x92490000;, dropping the low half set just before, and mov w0, #-1 (MOVN) to w0 = 0;. Keep MOVK's other bits and invert MOVN's immediate — claimed: htcom-code, branch bugfix/r3.11-a64-move-wide-immediates
  • Fold an AArch64 MOVZ/MOVK sequence into one constant in pseudo-C: mov w9, #0x2493 then movk w9, #0x9249, lsl #16 reads as two statements, not w9 = 0x92492493; — claimed: Jensen-Yao, branch bugfix/r3.11-a64-movz-movk-constants
  • Enforce stand-in per-body and total byte budgets on the text that is actually retained after redaction; deduplicate consecutive samples before charging the capture budget. — claimed: Afloat16, branch bugfix/r3.11-standin-redacted-budget-20261103b Revision d601bc4e passed full Linux CI: all four retained-budget regressions and six deduplication regressions passed.
  • Preserve every observation when comparing evidence bundles, including repeated operation/subject records and identical-record multiplicity. — claimed: Afloat16, branch bugfix/r3.11-bundle-record-multiplicity-20261110
  • Read only regular files for binary, text and evidence-bundle inputs, enforce whole-file byte caps while reading, and hash prefixes only after EOF instead of trusting a prior size report. — claimed: Afloat16, branch bugfix/r3.11-bounded-file-inputs-20261110
  • A text source (a script, Markdown) had no tool that reads it as text outside a build (#539): build_file_read works only while a project is being built, and the native tools stop at "Unknown file magic", while triage pointed to native_bytes_show, whose error points back to triage. Reproduced on main (0c1f7f7) with a .md and a .ps1. core_text_read (Deep only) reads it paged by lines with coverage, counts invalid bytes and cuts long lines; triage names it for a text file it doesn't otherwise recognise — claimed: chordpli, branch feature/r3.11-text-source-read
  • Keep text-read coverage partial when a line was cut, even when that line is the last one in the file. — claimed: Afloat16, branch bugfix/r3.11-cut-text-coverage-20261110
  • Live-check #539 in an agent chat (Codex, as reported, and Claude Code): the agent reaches core_text_read for a script or Markdown target, instead of searching bytes. On macOS with chat --deep and a .md or .ps1 target: Claude Code 2.1.322 (.md) called triage and core_text_read together; Codex 0.162.1 read the .ps1 with core_text_read straight away, and for the .md tried core_overview_show ("Unknown file magic"), then triage, then core_text_read. All three quoted the file exactly; none used a byte or string search — claimed: chordpli, branch test/r3.11-text-read-live-chat
  • A text file is still "A file we don't recognise yet" everywhere but triage's answer: the target's kind stays unknown, so core_targets_list and triage's description say so, and Codex tried core_overview_show on a .md before triage. Give text its own target kind (or description) so the list and the app call it text — claimed: Martin-Aziz, branch bugfix/r3.11-text-target-kind
  • AArch64 register and bitmask-immediate moves read their zero register: mov x19, x0 (ORR from xzr) lifts to x19 = xzr; and mov w0, #0xff00ff00 to w0 = wzr;. Move the last operand — claimed: htcom-code, branch bugfix/r3.11-a64-mov-alias-source
  • A folder walk (walk_dir) was bounded only by the files it kept: after max_files it went on through the whole tree, it read every entry of a folder before looking at a limit, and it kept every folder and link, so a tree of empty folders or links, or one huge folder, had no limit. WalkLimits bounds the entries examined and the path bytes held (from the file limit, 4,000,000 entries at most); one folder is listed through a bounded first-names-in-order selection, so a cut listing is the same every time; the walk stops at the first file or entry over a limit; cancel is checked while a folder is read. An ASAR's .unpacked folder and source_paths folders say when their listing was cut. The tree-of-links test skipped itself on the Windows account used (it can't create links); it needs a run on Linux or macOS — claimed: hichang4u, branch bugfix/r3.11-walk-limits
  • walk_dir still reads each .gitignore whole, whatever its size: cap it like other input reads and say when one was skipped — claimed: AlexJoaquimPereira, branch bugfix/r3.11-gitignore-read-cap
  • The walk's limits are chosen, not measured (4,000,000 entries at most, 256 path bytes an entry): time a walk at the ceiling on a real disk and measure what it holds
  • core_archives_compare reports rows_capped and a partial-coverage limitation when the added, removed, changed or moved lists are cut at max_rows, while its counts stay complete. — claimed: AlexJoaquimPereira, branch bugfix/r3.11-archive-compare-rows

R3.12 Format parser correctness

Owner: open · Status: in progress

  • Keep source-to-bundle comparisons uncertain when source-map inventories are truncated, preserving positive matches and reporting partial tool evidence. — claimed: Afloat16, branch bugfix/r3.12-source-map-comparison-uncertainty-20261111

  • Bound expanded source-map inventories by retained path bytes and entries before allocation, preserving bounded prefixes and reporting truncation. — claimed: Afloat16, branch bugfix/r3.12-source-inventory-budget-20261111

  • Preserve readable ZIP64 end-record fields with framing warnings and classic fallback for unusable locators; reject undersized, overflowing or truncated fixed records. — claimed: Afloat16, branch bugfix/r3.12-zip64-record-extent

  • Recover ZIP64 records shifted by prepended data with a bounded lookbehind and preserve the fallback for locator-shaped classic comments. — claimed: Afloat16, branch bugfix/r3.12-zip64-prefix-recovery-20261111

  • Preserve fractional seconds when projecting binary plist dates and keyed-archive NSDate values, including dates before the reference epoch and rounding across a day boundary. — claimed: Afloat16, branch bugfix/r3.12-plist-fractional-dates-20261111

  • Mark Python code inventories as cut short whenever a name or constant is shortened, and count truncation markers in the retained-text budget. — claimed: Afloat16, branch bugfix/r3.12-pyc-truncated-text-evidence-20261110

  • Keep worker, service-worker and utility-process entry modules in their own execution role instead of inheriting the parent role through spawn edges. — claimed: Afloat16, branch bugfix/r3.12-worker-role-boundaries

  • Read the explicit module-format byte in current Node SEA blobs while retaining the older layouts; report the stored format and recover the main script, assets and execution arguments at their real offsets. — claimed: Afloat16, branch bugfix/r3.12-node-sea-module-format-20261110

  • Deduplicate modules before queueing role propagation, so repeated imports do not exhaust the application graph's traversal limit before reachable modules receive their roles. — claimed: Afloat16, branch bugfix/r3.12-module-role-queue

  • Resolve webpack module-loader calls by the wrapper parameter's binding, so nested parameters and local variables with the same name do not create dependency edges. — claimed: Afloat16, branch bugfix/r3.12-webpack-require-bindings

  • An application graph counts the edges its cap drops, reports an edge_cap limitation, and js_app_map marks coverage partial when the graph was cut by its node or edge cap. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-app-graph-edge-cap

  • js_app resolves only the first 2,000 imports of a module and drops every ModuleFacts.truncated kind, so later dependencies lose their edge and role and no limitation says a list was cut; resolve every retained import, report facts truncation, and make js_app_map coverage partial. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-import-facts-truncation

  • Preserve webpack factory parameter names when direct eval can read them. — claimed: Plastic-time, branch bugfix/r3.12-js-eval-param-renaming-1115

  • Preserve void zero expressions when a with statement can dynamically resolve undefined. — claimed: Plastic-time, branch bugfix/r3.12-with-void-1116

  • Keep sequence splitting from introducing JavaScript strict-mode directives. — claimed: Plastic-time, branch rrfz/bugfix-r3.12-js-sequence-directives

  • Match numeric bundle module and chunk identities by their JavaScript value, so hexadecimal, separated and exponent literals connect to equivalent dependency references. — claimed: Afloat16, branch bugfix/r3.12-bundle-numeric-identities

  • Preserve scoped package names in JavaScript license banners. — claimed: Plastic-time, branch bugfix/r3.12-js-scoped-vendor-1115

  • Allocate distinct recovered module paths after repeated content-hash collisions. — claimed: Plastic-time, branch rrfz/bugfix-r3.12-js-module-collisions

  • Bound the text JavaScript facts build from constant strings, so repeated concatenation of a constant with itself cannot grow without limit. — claimed: Plastic-time, branch bugfix/r3.12-js-constant-text-budget

  • Read .cts files as CommonJS and .mts files as ES modules, as Node and TypeScript do, instead of guessing from the syntax. — claimed: Plastic-time, branch bugfix/r3.12-js-cts-commonjs

  • Reject physical line breaks in regular GraphQL strings, including after a backslash. — claimed: Plastic-time, branch bugfix/r3.12-graphql-string-line-breaks

  • Keep escaped triple quotes inside GraphQL block strings when reading operations. — claimed: Plastic-time, branch bugfix/r3.12-graphql-block-strings

  • Read + in JavaScript constant text as JavaScript does: numbers add ('/v' + (1 + 2) is /v3, not /v12) and number literals read as their value (0x10 is 16). — claimed: Plastic-time, branch bugfix/r3.12-js-numeric-addition

  • Walk past the old GNU sparse extension blocks to list the next TAR member. — claimed: Plastic-time, branch bugfix/r3.12-gnu-sparse-extensions

  • Accept the zero checksum placeholder on GNU-style CRC cpio symbolic links. — claimed: Plastic-time, branch bugfix/r3.12-cpio-symlink-crc

  • Refuse oversized GNU tar long names and link targets instead of silently truncating member identities. — claimed: Afloat16, branch bugfix/r3.12-tar-long-identities-20261109 Source de2c6390 passed formats tests, fmt and clippy and layers, locks, secret scanning, wasm checks and Cargo.lock stability. The unchanged-code baseline reproduces the long-name identity regression.

  • Refuse managed engine installation when GNU tar name metadata stops the archive walk, before writing files; retain partial listings for inspection and accept complete member data without an end marker. — claimed: Afloat16, branch bugfix/r3.12-tar-install-stop-reasons-20261109 Source 3e588c55 passed formats and engines tests, formatting, clippy and L0 gates. The unchanged-production baseline reproduces four failed guards while two compatibility controls pass.

  • Require the complete padded extent of GNU tar name and link records before allowing installation, including extents that overflow checked offsets; preserve ordinary member reads without trailing padding. — claimed: Afloat16, branch bugfix/r3.12-gnu-tar-padded-records-20261109 Exact source 25c45ef0 passed formats and engines tests (1,200 passed, one existing network test ignored), fmt, clippy and L0 gates. The unchanged-production baseline reproduces four failed guards while seven compatibility controls pass.

  • Extract legacy PyInstaller PYZ data as raw files and report modern namespace packages without fabricating bytecode. — claimed: Afloat16, branch bugfix/r3.12-pyz-member-kinds

  • Bound retained expanded PYZ paths across nested archives before allocating repeated prefixes; preserve complete entry identities and declared module counts in partial listings. — claimed: Afloat16, branch bugfix/r3.12-pyz-expanded-path-budget-20261110

  • Stop a PyInstaller CArchive listing at a member name without its required NUL terminator, preserving earlier readable members. — claimed: Plastic-time, branch bugfix/r3.12-pyinstaller-name-terminator-1111

  • Keep PyInstaller CArchive member reads before the table and cookie, including signed files with trailing bytes. — claimed: Afloat16, branch bugfix/r3.12-carchive-payload-range-20261109 Source a5bcfaed passed formats tests, fmt and clippy and layers, locks, secret scanning, wasm checks and Cargo.lock stability. The unchanged-code baseline reproduces the new regression cases.

  • Build only the requested keyed-archive object rows and count shared class names without copying them for every object. — claimed: Afloat16, branch bugfix/r3.12-keyed-archive-page-before-copy-20261110

  • Bound rejected ASAR names by the archive listing limit. — claimed: Afloat16, branch bugfix/r3.12-asar-limit-bounds-20261107

  • Bound retained ASAR entry and rejected-name paths before expanding shared directory prefixes; preserve complete member identities and report a partial listing when the byte budget is reached. Extract recovered members with partial coverage and no claimed total when a listing is incomplete. — claimed: Afloat16, branch bugfix/r3.12-asar-path-budget-20261110

  • Keep embedded Godot pack tables and member reads before the executable's pack trailer. — claimed: Afloat16, branch bugfix/r3.12-godot-trailer-boundary-20261109 Source 9744a0569a7c; format checks and compatibility gates passed.

  • Keep .NET bundle member reads before the manifest. — claimed: Afloat16, branch bugfix/r3.12-dotnet-data-bounds-20261107

  • Bound retained Node SEA asset and argument metadata while still validating the complete blob layout. — claimed: Afloat16, branch bugfix/r3.12-node-sea-retained-budget-20261109 Source ce34c2446499; format checks and compatibility gates passed.

  • Check the decoded byte count for packed files whose manifest declares their original size. — claimed: Afloat16, branch bugfix/r3.12-packed-member-size

  • Validate gzip packed-member trailers and decode concatenated members within one output budget. — claimed: Afloat16, branch bugfix/r3.12-packed-gzip-integrity-20261109 Source 3f3ec6a4 passed formats tests, fmt and clippy and layers, locks, secret scanning, wasm checks and Cargo.lock stability. The unchanged-code baseline reproduces the new regression cases.

  • Decode all frames of concatenated Zstandard members and account for their combined output. — claimed: Afloat16, branch bugfix/r3.12-zstandard-multiple-frames

  • Read concatenated XZ streams and four-byte stream padding while sharing one output budget. — claimed: Afloat16, branch feature/r3.12-xz-concatenation-20261109

  • Recover only the first XZ stream in a SquashFS block, ignoring trailing bytes without losing valid data. — claimed: Afloat16, branch bugfix/r3.12-squashfs-xz-first-stream-20261111

  • SQLite TEXT values cut at the 4 KiB cap can end in a spurious U+FFFD when the cut lands inside a character; cut at a UTF-8/UTF-16 boundary and keep the truncated flag — tests cover a UTF-8 character, a UTF-16 surrogate pair and a UTF-16 continuation-like byte at the cut; invalid bytes keep their lossy prefix. Claimed: AlexJoaquimPereira, branch bugfix/r3.12-sqlite-text-cap

  • WITHOUT ROWID rows stored before ALTER TABLE ADD COLUMN show the added columns' defaults, as rowid tables already do, instead of NULL. — claimed: Plastic-time, branch bugfix/r3.12-sqlite-without-rowid-added-columns

  • Discard decoded SQLite rows before the requested offset instead of retaining every preceding row, and keep zero-sized pages bounded while preserving readable-row offsets and key order. — claimed: Afloat16, branch bugfix/r3.12-sqlite-page-retention-20261110

  • Fill added SQLite columns whose DEFAULT is a hex integer or a literal in parentheses (DEFAULT 0x10, DEFAULT (5)); they show NULL with a note today. Hex integers keep SQLite's 64-bit two's-complement value; one parenthesized literal is filled, while expressions and nested parentheses stay unknown with the note. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-sqlite-added-defaults

  • Give filled-in SQLite defaults their column's affinity, as SQLite reads them: a TEXT column reads DEFAULT 5 as '5' and DEFAULT 007 as '7'; an INTEGER or NUMERIC column reads DEFAULT '7' as 7 and DEFAULT 5.0 as 5; a REAL column reads DEFAULT '1.5' as 1.5; a BLOB or untyped column, and an ANY column in a STRICT table, keep the literal as written. — claimed: Chr0mium24, branch bugfix/r3.12-sqlite-default-affinity

  • Fill added SQLite columns whose DEFAULT is a bare or double-quoted word (DEFAULT abc, DEFAULT "abc"), a signed string (DEFAULT -'5') or a number with digit separators (DEFAULT 1_000); they show NULL with a note today. Bare and double-quoted words are text constants; a signed numeric string and single digit separators take the column's affinity, and a separator-bearing REAL fills a missing TEXT field with its unseparated spelling. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-sqlite-default-literals

  • Read a TEXT column's REAL default as SQLite does: DEFAULT 1e3 and DEFAULT 5. are '1000.0' and '5.0' in SQLite. A non-integer REAL default now uses SQLite's REAL text form, including Inf/-Inf; integer literals keep integer text. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-sqlite-real-text

  • Preserve decimal literal spelling when filling a missing TEXT field from a SQLite default with digit separators. — claimed: Afloat16, branch bugfix/r3.12-sqlite-added-default-spellings

  • Read a signed quoted numeric default with SQLite's coercion before the column's affinity: DEFAULT -'1e3' is '-1000.0' on TEXT and -1000 on INTEGER, and DEFAULT -'1.50' is '-1.5'; integer text keeps integer formatting. The quoted text is coerced to an INTEGER or REAL before affinity, and TEXT's REAL value uses SQLite's REAL text form. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-sqlite-signed-quoted

  • Whole numbers in SQLite REAL columns read as REAL: SQLite stores a REAL with no fractional part as an integer and turns it back into a REAL when it reads the column, so a REAL, FLOAT or DOUBLE column holding 1.0 shows 1.0 instead of 1. — claimed: Chr0mium24, branch bugfix/r3.12-sqlite-real-affinity

  • Require a complete DNS question's type and class fields before retaining it, and keep truncated questions out of packet inventories while reporting the incomplete message — claimed: Afloat16, branch bugfix/r3.12-dns-question-class-review-20261104

  • Preserve DNS wire-label identity and require valid HTTP start lines before reporting packet evidence. — claimed: Afloat16, branch bugfix/r3.12-capture-evidence

  • Reject impossible IPv4 header and total lengths. — claimed: Afloat16, branch bugfix/r3.12-ipv4-header-lengths-20261105

  • Validate the version in IPv6 frames. — claimed: Afloat16, branch bugfix/r3.12-ipv6-version-20261105

  • Decode VLAN-encapsulated network packets in Linux cooked v1 and v2 captures. — claimed: Afloat16, branch bugfix/r3.12-cooked-vlan-packets

  • Reject negative GNU base-256 values in unsigned tar header fields. — claimed: Afloat16, branch bugfix/r3.12-tar-octal-fields

  • Recognize GNU sparse PAX metadata and report unsupported reconstruction instead of returning packed extents as complete files. — claimed: Afloat16, branch bugfix/r3.12-tar-pax-sparse-20261109 Source daaeb4fe passed formats tests, fmt and clippy and layers, locks, secret scanning, wasm checks and Cargo.lock stability. The unchanged-code baseline reproduces the new regression cases.

  • Recognize ZIP directories from DOS directory attributes and Unix file-type metadata when the entry name has no trailing slash. — claimed: Afloat16, branch bugfix/r3.12-zip-directory-attributes

  • Read ZIP bzip2 and XZ members through the existing bounded pure-Rust decompressors. — claimed: Afloat16, branch feature/r3.12-zip-bzip2-xz-20261109 Source 4cac3638 passed formats tests, fmt and clippy and layers, locks, secret scanning, wasm checks and Cargo.lock stability. The unchanged-code baseline reproduces the new regression cases.

  • Use a verified Info-ZIP Unicode Path extra field when listing a member stored under a legacy filename. — claimed: Afloat16, branch feature/r3.12-zip-unicode-path

  • Decode ZIP method 14 property headers and size or end-marker framing with bounded LZMA output. — claimed: Afloat16, branch feature/r3.12-zip-lzma-20261109 Source 61fdab39 passed formats tests, fmt and clippy and layers, locks, secret scanning, wasm checks and Cargo.lock stability. The unchanged-code baseline reproduces the new regression cases.

  • Report ECE and CWR alongside the existing TCP flags in packet inventories. — claimed: Plastic-time, branch bugfix/r3.12-tcp-ecn-flags

  • Stop decoding extension headers inside noninitial IPv6 fragments. — claimed: Afloat16, branch bugfix/r3.12-ipv6-fragment-payload-20261105

  • Enforce the expanded DNS name wire-length limit. — claimed: Afloat16, branch bugfix/r3.12-dns-name-wire-length-20261105

  • Report truncated DNS TXT strings as incomplete. — claimed: Afloat16, branch bugfix/r3.12-dns-txt-truncation-20261105 Preserve empty ANY/NONE UPDATE prerequisites with zero TTL; keep TXT validation scoped to parsed answers/prerequisites. Preserve empty multicast/unicast DNS-SD TXT records using wire-label service instance names.

  • Verify ECDSA keys using their declared named curve and reject missing, malformed or unsupported parameters instead of selecting a curve from the point length. — claimed: Afloat16, branch bugfix/r3.12-ec-named-curve-20261110

  • Verify RSASSA-PSS signatures using their declared digest, mask and salt parameters, enforce PSS public-key constraints, and require signed and outer certificate parameters to agree. — claimed: Afloat16, branch bugfix/r3.12-rsa-pss-parameters-20261110

  • Retain readable ASN.1 elements with redundant length encodings, mark non-minimal lengths, and report the not-strict-DER limitation in certificate and signature views. — claimed: Afloat16, branch bugfix/r3.12-canonical-der-lengths

  • Validate DER algorithm identifiers, signature and key bit strings, and name fields before using them as certificate verification inputs. — claimed: Afloat16, branch bugfix/r3.12-der-verification-fields

  • Reject certificates whose signed signature-algorithm OID conflicts with the outer algorithm before reporting a verified signature. — claimed: Afloat16, branch bugfix/r3.12-certificate-signature-oids-20261110

  • Decode complete BMPString and UniversalString values, legacy TeletexString as Latin-1 and the supported ASCII name strings; retain malformed or unknown values as hexadecimal DER instead of lossy text. — claimed: Afloat16, branch bugfix/r3.12-certificate-name-identity

  • Distinguish literal punctuation and multi-valued attributes in the displayed X.509 issuer and subject. — claimed: Afloat16, branch bugfix/r3.12-certificate-name-rendering

  • Match CMS signer certificates by issuer and serial number or subject key identifier; leave absent matches unknown instead of selecting an unrelated certificate and chain. — claimed: Afloat16, branch bugfix/r3.12-cms-signer-identities-20261110

  • Read HTTP Host fields only from the header section. — claimed: Afloat16, branch bugfix/r3.12-http-host-boundary-20261105

  • Select certificate issuers by successful signature verification among matching names, follow self-issued key rollover, and report interrupted chains as incomplete. — claimed: Afloat16, branch bugfix/r3.12-certificate-issuer-selection-20261110

  • Distinguish certificate-chain cycles from the 16-link walk limit when explaining why issuer verification stopped (follow-up to #1518). — claimed: Afloat16, branch bugfix/r3.12-certificate-cycle-note-20261110

  • Keep capture flow timestamps in chronological order. — claimed: Afloat16, branch bugfix/r3.12-capture-flow-time-range-20261105

  • Reject incomplete simple packet block payloads. — claimed: Afloat16, branch bugfix/r3.12-pcapng-simple-block-length-20261105 Count rejected lengths against the packet scan limit and summarize them once at the end of the walk.

  • Bound rejected packet records, validate timestamp fractions and check pcapng section-local interfaces. — claimed: Afloat16, branch bugfix/r3.12-capture-timing-framing

  • Report capped ZIP, tar and cpio member walks as incomplete firmware extraction, even when skipped directory entries hide later files. — claimed: Afloat16, branch bugfix/r3.12-firmware-member-scan-completeness-20261110

  • Recognize Unix file modes from macOS ZIP hosts and the DOS directory attribute from VFAT hosts, preserving directory and symlink identity without a trailing slash. — claimed: Afloat16, branch bugfix/r3.12-macos-vfat-directory-attributes

  • Keep ZIP directory entries within the declared directory. — claimed: Afloat16, branch bugfix/r3.12-zip-directory-bounds-20261105

  • Refuse split ZIP archives without their other volumes. — claimed: Afloat16, branch bugfix/r3.12-zip-multidisk-20261105

  • Check Unity bundle blocks against their declared output lengths. — claimed: Afloat16, branch bugfix/r3.12-unity-bundle-decompressed-length-20261105

  • Read only requested byte ranges from uncompressed Unity bundle blocks so small metadata and asset reads do not allocate or reject entire stored blocks, while preserving size checks and compressed-block caching. — claimed: Afloat16, branch bugfix/r3.12-unity-stored-block-ranges-20261110

  • Apply standalone LZMA output budgets to its internal dictionary and reject oversized declared output before decoding. — claimed: Afloat16, branch bugfix/r3.12-standalone-lzma-budget-20261109 Source 5da8880d passed formats tests, fmt and clippy and layers, locks, secret scanning, wasm checks and Cargo.lock stability. The unchanged-code baseline reproduces the new regression cases.

  • Expand repeated LZ4 match bytes in blocks when reading Unity bundles. — claimed: Afloat16, branch refactor/r3.12-expand-lz4-matches

  • Require terminated cpio pathnames and a complete, empty trailer before reporting an archive as complete. — claimed: Afloat16, branch bugfix/r3.12-cpio-path-and-trailer

  • Identify regular files sharing an inode and device in newc and CRC archives so hard-link aliases are visible in listings. — claimed: Afloat16, branch feature/r3.12-cpio-hard-link-evidence

  • Locate Unity bundle metadata within the declared bundle size. — claimed: Afloat16, branch bugfix/r3.12-unity-bundle-declared-size-20261105

  • Check that Unity bundle compressed blocks end before the trailing block table, or the declared bundle end for front-positioned tables, so bytes outside the data region cannot supply block contents. — claimed: Codex, branch bugfix/r3.12-unity-bundle-data-region; PR #1021

  • Check Unity bundle member ranges against the complete uncompressed block stream, retain invalid entries as unreadable with explicit listing warnings, and preserve valid sibling members, cross-block ranges and empty entries. — claimed: Afloat16, branch bugfix/r3.12-unity-node-ranges-20261110

  • Validate Unity serialized metadata offsets and lengths before reading so short inputs cannot panic or borrow object bytes. — claimed: Afloat16, branch bugfix/r3.12-unity-metadata-range-20261109 Source de9d7444f509; format checks and compatibility gates passed.

  • Reject Unity serialized objects outside their declared file. — claimed: Afloat16, branch bugfix/r3.12-unity-serialized-object-ranges-20261105

  • Count WebAssembly locals across all declaration groups in retained function bodies. — claimed: Afloat16, branch bugfix/r3.12-wasm-locals

  • Preserve compression method slots in Unreal pak indexes. — claimed: Afloat16, branch bugfix/r3.12-unreal-pak-compression-slots-20261105

  • Require complete terminators in Unreal FString values. — claimed: Afloat16, branch bugfix/r3.12-unreal-fstring-terminators-20261105

  • Count every compression block spanned by an IoStore chunk. — claimed: Afloat16, branch bugfix/r3.12-iostore-unaligned-chunk-blocks-20261105

  • Reject IoStore directory rows that name a missing chunk or name instead of reporting a zero-byte file or fabricated path. — claimed: Afloat16, branch bugfix/r3.12-iostore-directory-references

  • Reject truncated IoStore fixed headers. — claimed: Afloat16, branch bugfix/r3.12-iostore-complete-header-20261105

  • Reject CBOR collections that exceed the EVM metadata reader's limits and invalid text strings. — claimed: Afloat16, branch bugfix/r3.12-evm-cbor

  • Keep ordinary EVM metadata arrays as arrays, formatting three-part versions only for compiler version fields. — claimed: Plastic-time, branch bugfix/r3.12-evm-metadata-array-values-1111

  • Keep EVM bytecode intact when its CBOR trailer repeats a metadata key or uses a non-text map key. — claimed: Afloat16, branch bugfix/r3.12-evm-cbor-map-keys-20261110

  • Require the selector division operand order, the Vyper memory load and the four-byte calldata threshold before reporting those dispatch patterns; invalid jump destinations retain unknown payability. — claimed: Afloat16, branch bugfix/r3.12-contract-dispatch-evidence

  • Require evidence of a rejecting payment branch before labelling EVM functions nonpayable; reading CALLVALUE alone accepts payments. — claimed: Afloat16, branch bugfix/r3.12-evm-payment-guards

  • Avoid rescanning every retained EVM selector for repeated dispatcher comparisons. — claimed: Afloat16, branch bugfix/r3.12-evm-selector-dedup-20261110

  • Reject invalid marshal long integer digits. — claimed: Afloat16, branch bugfix/r3.12-pyc-long-limb-validation-20261105

  • Check expanded RGBA image sizes and enforce output budgets before allocation. — claimed: Afloat16, branch bugfix/r3.12-pixel-rgba-output-overflow-20261105

  • Reject duplicate keys in safetensors headers before they silently replace tensor descriptions or model metadata. — claimed: Afloat16, branch bugfix/r3.12-safetensors-duplicate-keys-20261110

  • Report malformed safetensors metadata, element widths and payload coverage. — claimed: Afloat16, branch bugfix/r3.12-safetensors-layout

  • Disclose incomplete asset catalog tree reads instead of reporting silently empty complete listings. — claimed: Afloat16, branch bugfix/r3.12-car-tree-completeness-20261110

  • Read Apple BOM block indices and named variables only within the directory byte ranges declared by the container. — claimed: Afloat16, branch bugfix/r3.12-bom-directory-bounds

  • Count empty tensor shapes consistently across axis orders. — claimed: Afloat16, branch bugfix/r3.12-tensor-zero-dimension-product-20261105

  • Read certificate choices from the SignedData certificate set only and enforce the certificate limit across the blob. — claimed: Afloat16, branch bugfix/r3.12-signeddata-certificate-set

  • Find APK signatures through valid ZIP comments and framed central directories. — claimed: Afloat16, branch bugfix/r3.12-apk-sign-eocd-comment-boundary-20261105

  • Require APK signer SDK ranges and ordered signature-algorithm lists to match their signed copies, with complete required signer fields, before reporting verification. — claimed: Afloat16, branch bugfix/r3.12-apk-signed-metadata-20261110

  • Include an Android package's type-ID offset in resource identities while retaining its local type-name indices; reject shifted IDs outside the one-byte type field. — claimed: Afloat16, branch bugfix/r3.12-arsc-type-id-offset-20261110

  • Keep Android resource entry data after the offset table; retain readable identities with a warning when bag headers or records are incomplete, without inventing parent or bag values. — claimed: Afloat16, branch bugfix/r3.12-resource-table-identity

  • Keep Android resource packages, child chunks and library records inside their enclosing declared ranges. — claimed: Afloat16, branch bugfix/r3.12-arsc-parent-chunk-boundaries-20261109 Source ea680eeb9e92; format checks and compatibility gates passed.

  • Retain carrier zero, grammatical gender, round-screen and color capabilities, and keyboard/navigation availability in resource configuration names. — claimed: Afloat16, branch bugfix/r3.12-resource-device-variants

  • Keep explicit script and variant subtags when grouping Android resource strings by locale. — claimed: Afloat16, branch bugfix/r3.12-android-script-locale

  • Bound decoded Android string-pool bytes and disclose partial pools in binary XML and resource tables, so shared offsets cannot multiply a small chunk into unbounded retained strings. — claimed: Afloat16, branch bugfix/r3.12-android-pool-retained-bytes-20261110

  • Carry incomplete Android manifest decoding into package and component evidence coverage, including the last component page, without substituting ZIP-entry counts for document completeness. — claimed: Afloat16, branch bugfix/r3.12-android-manifest-coverage-20261110

  • Read 16-bit native pointers without consuming adjacent bytes. — claimed: Afloat16, branch bugfix/r3.12-binary-pointer

  • Keep same-named imported functions attached to their own library and preserve an explicitly recorded pointer slot. — claimed: Afloat16, branch bugfix/r3.12-import-library-slots-20261109

  • Preserve a unique versioned ELF import slot when relocation records omit the library name (follow-up to #1365). — claimed: Afloat16, branch bugfix/r3.12-elf-versioned-import-slot-fallback-20261109 — completed: 1179870c55

  • Keep Mach-O load-command facts within their command. — claimed: Afloat16, branch bugfix/r3.12-macho-load-command-framing-20261105

  • Resolve Mach-O UNIXTHREAD entry points from a complete state record for the image CPU, including later state records. — claimed: Afloat16, branch bugfix/r3.12-macho-thread-state-20261109

  • Prevent Mach-O zero-fill sections from exposing unrelated file bytes during section lookup. — claimed: Afloat16, branch bugfix/r3.12-macho-zerofill-section-data

  • Keep native address translations and reads inside both the mapped range and its file-backed bytes. — claimed: Afloat16, branch bugfix/r3.12-binary-padding

  • native_strings_find lists at most 32 references per string without a count or flag, though its search is described as complete; add the reference count (complete within the analysis unless its xref limit was reached), a capped flag, the analysis xref-limit limitation and a display-cap limitation, as native_strings_browse already reports a count. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-string-reference-count

  • List each Objective-C class and category address once when ordinary and non-lazy registration tables overlap, keeping distinct same-named definitions separate. — claimed: Afloat16, branch bugfix/r3.12-objc-nonlazy-identity-20261111

  • Require complete Objective-C method record strides before reading absolute or relative method fields. — claimed: Afloat16, branch bugfix/r3.12-objc-method-record-stride-20261109

  • Keep Mach-O function starts within their declared payload. — claimed: Afloat16, branch bugfix/r3.12-macho-function-starts-payload-boundary-20261105

  • Apply one indirect-import budget across all Mach-O sections and report when repeated table references exhaust it. — claimed: Afloat16, branch bugfix/r3.12-macho-indirect-global-budget-20261109

  • A Mach-O executable's entry point comes from the segment whose file bytes hold its LC_MAIN offset. A segment with no file bytes, such as a __DATA that holds only zero-fill sections, records file offset 0 like __TEXT, and the entry was placed in that segment instead of in __TEXT (/bin/cat reads 0x100008588 for 0x100000588). — claimed: Chr0mium24, branch bugfix/r3.12-macho-entry-zero-fill

  • Name DriverKit and visionOS simulator Mach-O binaries in the binary view: its platform table stopped at 9 and 11, so a .dext read "unknown" while the dependency view said "driverkit". — claimed: Chr0mium24, branch bugfix/r3.12-macho-platform-names

  • Swift symbols with a protocol type decode. In a protocol list (…_p) each protocol is a context and a name with no P kind letter, but the demangler expected a type there, so an existential of a protocol named this way was reported invalid ("expected a type"): _$sSS6encode2toys7Encoder_p_tKF is Swift.String.encode(to: Swift.Encoder) throws -> (). — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-existentials

  • Swift function type attributes decode. Besides Ya (async), a function type can carry Yb (@Sendable), Yc (a global actor, with its type), YA (@isolated(any)) and YK (the error type of throws(E)), but the demangler read only Ya and reported the rest invalid ("unsupported function attribute"): _$s1x2v1yyYbcvp is x.v1 : @Sendable () -> (). — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-function-attributes

  • Swift value witness symbols decode. A symbol ending w and two letters is one of the functions the runtime calls to copy, move, destroy or read the enum tag of a value of a type, but the demangler had no w operator and reported it invalid ("unsupported operator 'w'"): _$ss6ResultOwug is getEnumTag value witness for Swift.Result. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-value-witnesses

  • Swift lazy witness table cache variables decode. A symbol ending WL names a type and the conformance it is used with, as Wl does, but the demangler read only one operand and reported it invalid ("N nodes left after decoding"): _$s7SwiftUI4EdgeO3SetVAESQAAWL is lazy protocol witness table cache variable for type SwiftUI.Edge.Set and conformance SwiftUI.Edge.Set : Swift.Equatable in SwiftUI. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-witness-cache

  • Swift enum case symbols decode. A symbol ending WC is the function that builds an enum case, but the demangler read WC as an unknown witness kind and reported it invalid ("unsupported witness kind"): _$s10FindMyBase3BitO3oneyA2CmFWC is enum case for FindMyBase.Bit.one(FindMyBase.Bit.Type) -> FindMyBase.Bit. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-enum-cases

  • Swift symbols decode opaque result types. A function, variable or subscript declared to return some P mangles that result as Qr (QR for a second or later one in a tuple), and a use of it elsewhere as QO and Qo, but the demangler reported all of them invalid ("unsupported associated type"): _$s1a2mkQryF is a.mk() -> some. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-opaque-types

  • Swift symbols decode private declarations. A private or fileprivate declaration's name is followed by an identifier for its file and LL, and the demangler reported every such symbol invalid ("unsupported operator 'L'"): _$s10DVTIconKit12_isDebugging33_8F51EEEBC836B21F9ED44F4C6D38A8CBLLSbvp is DVTIconKit.(_isDebugging in _8F51EEEBC836B21F9ED44F4C6D38A8CB) : Swift.Bool. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-private-names

  • Swift symbols decode outlined value operations. The demangler reported every WO and a letter (WOh, WOc, WOi0_) as "unsupported witness kind", so _$s10ACPToolLib11ResetSchemaVWOh should read "outlined destroy of ACPToolLib.ResetSchema", as swift-demangle prints it. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-outlined-ops

  • Swift symbols name the right type after an optional, an associated type or an extension. The demangler's substitution list left out Sg optionals and Qz/Qy associated types and added extensions, so a later A… reference named the wrong node ([Swift.String : Swift.String] where the symbol says Swift.String?) or fell out of range. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-substitutions

  • Charge saved Swift substitution trees and repeated copies against the expansion budget before cloning. — claimed: Afloat16, branch bugfix/r3.12-swift-substitution-copy-budget-20261110

  • Swift symbols read SQ as Swift.Equatable. The demangler's table of standard types mapped it to the enum Swift.ImplicitlyUnwrappedOptional, so _$sSQMp read "protocol descriptor for Swift.ImplicitlyUnwrappedOptional", a requirement read A: Swift.ImplicitlyUnwrappedOptional, and Equatable conformance descriptors and witness tables (…SQAAMc, …SQsWP) were reported invalid ("expected a protocol"). — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-equatable

  • Swift symbols spell the concurrency types as Sc and a letter. The demangler reported all of them invalid ("concurrency standard types are not supported"), so _$sScAMp should read "protocol descriptor for Swift.Actor" and ScT, ScG, ScS and the rest their own types, with an unknown letter after Sc still invalid. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-concurrency-types

  • Swift symbols name the type metadata singleton initialization cache. The demangler read Ml with the name of ML, so _$s10RealityKit30__RKEntityTriggerSpecificationOMl read "lazy cache variable for type metadata for RealityKit.__RKEntityTriggerSpecification" where swift-demangle prints type metadata singleton initialization cache for …. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-singleton-cache

  • Swift one-time initialization symbols name the variable. WZ and Wz were invalid; they now read "one-time initialization function for" and "one-time initialization token for" followed by the variable's name, as swift-demangle prints them. For example _$s10ACPToolLib11RemoveTokenV13configuration_WZ reads "one-time initialization function for configuration". — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-once-init

  • Require a complete Cargo package version before a registry or vendor path becomes Rust dependency evidence. — claimed: Afloat16, branch bugfix/r3.12-rust-dependency-versions

  • Swift symbols name the right generic parameter in an associated type. Qy numbers its parameter the way q does (Qy_ is the second parameter B, Qy0_ the third), but the demangler read Qy_ as A, so Swift.Zip2Sequence.Iterator.next() returned (A.Element, A.Element)? where the symbol says (A.Element, B.Element)?. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-associated-type-index

  • Swift metadata symbols name their kind the way swift-demangle does. MN reads as noncanonical specialized generic type metadata (_$sSRys6UInt16VGMN), Mj as an opaque type descriptor accessor key and Mz as a flag for loading canonical specialized generic type metadata. The demangler printed type metadata for, module descriptor and type metadata heap reference for for them. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-metadata-names

  • Decode binary plist markers and dictionary keys according to their stored types. — claimed: Afloat16, branch bugfix/r3.12-bplist-typed-objects

  • Swift initialisers of structs, enums, protocols and extensions read init. The demangler printed every allocating initialiser (fC) as __allocating_init, a name swift-demangle keeps for a class's own initialisers, so _$sSdyS2dcfC read Swift.Double.__allocating_init(Swift.Double) -> Swift.Double for Swift.Double.init(Swift.Double) -> Swift.Double. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-init-names

  • Swift operator functions decode. A function named by an operator is written as the operator's letters and oi (infix), op (prefix) or oP (postfix), and the demangler reported every one invalid (unsupported operator 'o'), so _$s10Foundation4DateV2eeoiySbAC_ACtFZ now reads static Foundation.Date.== infix(Foundation.Date, Foundation.Date) -> Swift.Bool. Private operators, and the witnesses, thunks and descriptors around an operator function, decode too. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-operator-names

  • Swift arguments without a label read _: when another argument of the same function, initialiser or subscript has one. The demangler printed the bare type, so _$s10Foundation13__DataStorageC6append_6lengthySV_SitF read Foundation.__DataStorage.append(Swift.UnsafeRawPointer, length: Swift.Int) -> () where swift-demangle prints append(_: Swift.UnsafeRawPointer, length: Swift.Int) -> (). — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-unlabeled-args

  • Swift witness table accessors decode. The demangler read Wl, Wb, WT and WI as unknown witness kinds and reported them invalid ("unsupported witness kind"): _$s7SwiftUI4EdgeO3SetVAESQAAWl is lazy protocol witness table accessor for type SwiftUI.Edge.Set and conformance SwiftUI.Edge.Set : Swift.Equatable in SwiftUI. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-witness-accessors

  • Apply the extracted-string limit in file-offset order across UTF-16LE and narrow strings, so an early wide string is not hidden by later ASCII. — claimed: Afloat16, branch bugfix/r3.12-earliest-extracted-strings

  • Keep quoted greater-than signs inside XML plist attributes instead of ending the tag early. — claimed: Plastic-time, branch bugfix/r3.12-plist-quoted-tag-delimiters-1112

  • Visit XML plist values only inside a complete, correctly encoded document. — claimed: Afloat16, branch bugfix/r3.12-xml-plist-document

  • Normalize literal XML line endings before decoding plist text, keeping character references intact. — claimed: Plastic-time, branch bugfix/r3.12-plist-xml-line-endings-1111

  • Count every JavaScript line terminator in fact locations. — claimed: Plastic-time, branch bugfix/r3.12-js-line-index-1115

  • Decode legacy octal escapes, Unicode line continuations and zero-padded braced Unicode escapes in JavaScript strings. — claimed: Plastic-time, branch bugfix/r3.12-string-escapes-1116

  • End JavaScript line comments at every language line terminator, preserving following tokens for the nesting admission guard and lexer fallback. — claimed: Afloat16, branch bugfix/r3.12-js-line-comment-boundaries

  • Count call, index and tagged-template chains and consecutive labels in the JavaScript nesting estimate, so sources nested deeper than the parser's stack go to the lexer fallback instead of overflowing it. — claimed: Plastic-time, branch bugfix/r3.12-js-postfix-chain-nesting

  • Mark the function an export { f } list names even when the list comes before the declaration, and only the module-level binding it names, not other functions called f. — claimed: Plastic-time, branch bugfix/r3.12-js-export-before-declaration

  • Count the remaining JavaScript nesting shapes that overflow the parser's stack with two million links: an if(a);else if(a);else … chain keeps the depth its ; clears, a class after extends nests a heritage, a run of prefix ++ counts, and / after an update operator, a member name or yield divides, so brackets after it are counted instead of hidden in a false regex; such sources go to the lexer fallback instead of the parser. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-js-nesting-overflow-shapes

  • Read a one-module webpack chunk only when its module is a factory function and the file names webpack or pushes onto (g = g || []), so an ordinary queue.push([[1], { callback: () => 42 }]) is not a webpack chunk. — claimed: Plastic-time, branch bugfix/r3.12-js-single-module-chunk

  • Skip GraphQL header comments before reading operation names and selection sets. — claimed: Plastic-time, branch bugfix/r3.12-graphql-comments-1116

  • Read only a URL's path for GraphQL and tRPC endpoints: the host ends at /, ? or #, so https://example.test?next=/graphql is not a GraphQL endpoint. — claimed: Plastic-time, branch bugfix/r3.12-js-url-query-api

  • Read and edit JSONC registrations whose line comments end at a bare carriage return. — claimed: Afloat16, branch bugfix/r3.12-jsonc-carriage-return-comments

  • Name code-directory flags with the bits Apple uses: hard 0x100, kill 0x200, expires 0x400, restrict 0x800, enforcement 0x1000, library-validation 0x2000. The table used 0x4–0x80 for them and named 0x200 and 0x800 as other flags, so a library-validation signature (0x2000) showed no flag names and 0x12200 read "allowed-entitlements,runtime". — claimed: Chr0mium24, branch bugfix/r3.12-codesign-flag-names

  • Keep code-signature children and verification reads within their declared superblob. — claimed: Afloat16, branch bugfix/r3.12-codesign-superblob-length-boundary-20261105

  • Require complete, single-valued CMS message-digest and content-type signed attributes with the specified tags, and bind the content type to the encapsulated content before reporting authentication. — claimed: Afloat16, branch bugfix/r3.12-cms-signed-attributes-20261110

  • Reject out-of-range or unterminated library names in the ELF dynamic string table. — claimed: Afloat16, branch bugfix/r3.12-elf-dynamic-names

  • Recover dynamic symbols through PT_DYNAMIC when an ELF retains other section headers but has no usable dynamic-symbol section. — claimed: Afloat16, branch bugfix/r3.12-elf-partial-section-symbols-20261109

  • Refuse short ELF64 header and section records and handle section-table offsets at the address limit safely. — claimed: Afloat16, branch bugfix/r3.12-sbpf-records

  • Keep ELF version definitions, requirements and indices within their declared section byte ranges. — claimed: Afloat16, branch bugfix/r3.12-elf-version-section-bounds-20261109

  • Keep PE optional-header reads within the declared header. — claimed: Afloat16, branch bugfix/r3.12-pe-optional-header-framing-20261105

  • pe_is_managed reads data-directory 14 at a fixed offset without checking the declared optional-header size, so a PE whose header ends before that directory can be classified as managed; bound it as pe::headers does. — claimed: Plastic-time, branch bugfix/r3.12-managed-header-bounds-1113

  • dotnet::pe::parse_pe reads NumberOfRvaAndSizes and data directory 14 at fixed offsets without checking the declared optional-header size, so a PE whose header ends before that directory can be parsed as a .NET assembly from section-table bytes, or refused with a misleading section error; bound it as pe::headers does. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-dotnet-pe-optional-header

  • The firmware carver's PE region reads the certificate directory and SizeOfHeaders at fixed offsets without checking the declared optional-header size, so a short header can over-carve into the next embedded object; bound them as pe::headers does. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-firmware-pe-certificate-dir

  • A valid ASAR header written as { "files": ... (a space) or with another key first is not recognized by the format-dispatch looks_like_asar, though the ASAR reader accepts it; use one predicate. — claimed: NicolasBohn, branch bugfix/r3.12-one-asar-predicate

  • Bound import descriptor walks by mapped file bytes and report incomplete metadata. Normal imports keep following the terminator when their declared size is short or zero; delay-load and bound metadata also respect their directory sizes. Preserve the shared retained-text budget. — claimed: Afloat16, branch bugfix/r3.12-pe-import-directory-bounds-20261109

  • Require bound PE module and forwarder names to include their NUL terminator inside the declared directory. — claimed: Afloat16, branch bugfix/r3.12-bound-import-terminators-20261111

  • Use the PE image's declared header extent when an RVA is not mapped by a section. — claimed: Afloat16, branch bugfix/r3.12-pe-header-rva-span

  • Keep PE entry points, import slots and exception function starts from wrapping image-base arithmetic into unrelated addresses. — claimed: Afloat16, branch bugfix/r3.12-pe-virtual-address-overflow-20261109

  • Keep relocation symbol indices inside a linked symbol table and require a linked string table before reporting a syscall name. — claimed: Afloat16, branch bugfix/r3.12-sbpf-symbols

  • Keep PE resource metadata within the resource directory. — claimed: Afloat16, branch bugfix/r3.12-pe-resource-directory-boundary-20261105

  • Bound PE export-table work and returned aliases even when function slots are empty or one function has many names. — claimed: Afloat16, branch bugfix/r3.12-pe-export-work-budget-20261109

  • Keep PE version strings whose value is empty (stored with no value at all), such as OleSelfRegister, whose presence is the whole point. — claimed: Plastic-time, branch bugfix/r3.12-pe-empty-version-strings

  • Keep fixed PE file and product versions inside a complete declared VS_FIXEDFILEINFO value and its enclosing version node. — claimed: Afloat16, branch bugfix/r3.12-pe-fixed-version-bounds-20261110

  • Read PE version strings stored as binary (wType 0), which some linkers write; they're dropped today (FileDescription in pthreadVC2.dll). A binary leaf at the string depth is decoded as UTF-16LE; a binary node with children stays a container. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-pe-binary-version-strings

  • Decode inline source-map data URLs without treating their URL fragments as payload; recognize case-insensitive scheme and base64 markers. — claimed: Afloat16, branch bugfix/r3.12-source-map-data-url-fragment

  • Apply WebAssembly name and producer inventory limits across repeated custom sections and report truncation. — claimed: Afloat16, branch bugfix/r3.12-wasm-custom-inventory-budget-20261109

  • List top-level WebAssembly component sections without counting embedded module sections as component sections. — claimed: Plastic-time, branch bugfix/r3.12-wasm-component-layout

  • Bound bytes retained when WebAssembly types and function names are expanded through references, reuse formatted type signatures, and disclose omitted derived metadata. — claimed: Afloat16, branch bugfix/r3.12-wasm-reference-bytes-20261110 Source 232100030ace; format checks and compatibility gates passed.

  • Report WebAssembly metadata truncation accurately in summary and complete tool views, including byte-budget exhaustion before entry limits (follow-up to #1540). — claimed: Afloat16, branch bugfix/r3.12-wasm-metadata-limits-20261110

  • Accept only valid WebAssembly names when reporting custom-section debug and source map URLs. — claimed: Afloat16, branch bugfix/r3.12-wasm-debug-names

  • Find source map annotations after every ECMAScript line terminator. — claimed: Plastic-time, branch bugfix/r3.12-sourcemap-line-endings

  • Read source map hints inside complete trailing block comments with either annotation marker. — claimed: Plastic-time, branch bugfix/r3.12-sourcemap-block-boundaries

  • Decode the full signed 32-bit source map VLQ range and refuse mapping fields outside it. — claimed: Afloat16, branch bugfix/r3.12-sourcemap-vlq-range

  • Reject source map mappings with missing source or name entries. — claimed: Plastic-time, branch rrfz/bugfix-r3.12-mapping-index-20261110

  • Group Go functions by the declaring package when a generic argument names another import path. — claimed: Afloat16, branch bugfix/r3.12-go-generic-owner

  • Keep Go function tables and names within their section. — claimed: Afloat16, branch bugfix/r3.12-go-pclntab-section-boundary-20261105

  • Preserve complete Go functions before a damaged later table entry and report partial coverage when the scan stops at structural damage or its unreadable-name limit. — claimed: Afloat16, branch bugfix/r3.12-go-pclntab-prefix-20261110

  • Bound retained Go function-name bytes when pclntab records share name references, preserving recovered functions and explicitly disclosing omitted names. — claimed: Afloat16, branch bugfix/r3.12-go-name-bytes-20261110

  • Keep the report's function display limit from marking a valid large WebAssembly module as mismatched. — claimed: Afloat16, branch bugfix/r3.12-wasm-function-count

  • Retain decoded CLI signatures with trailing data and report the unused bytes as a limitation, while keeping nested function-pointer signatures incremental. — claimed: Afloat16, branch bugfix/r3.12-cli-signature-consumption-20261111

  • Preserve warnings from every view used by the managed application graph, deduplicating diagnostics before the display limit. — claimed: Afloat16, branch bugfix/r3.12-managed-graph-warnings-20261111

  • Reject TypeSpec handles where CLI class/value-type and generic signatures require TypeDef or TypeRef, while retaining standalone TypeSpec and custom-modifier decoding. — claimed: Afloat16, branch bugfix/r3.12-dotnet-signature-type-handles-20261110

  • Keep CLI multidimensional array ranks and bounds distinct from vector signatures. — claimed: Afloat16, branch bugfix/r3.12-cli-array-shapes-20261111

  • Recognize volatile type notation from required System.Runtime.CompilerServices.IsVolatile modifiers across the complete modifier list. — claimed: Afloat16, branch bugfix/r3.12-dotnet-volatile-modifiers-20261110

  • Preserve complete Unicode characters in shortened .NET strings. — claimed: Afloat16, branch bugfix/r3.12-dotnet-user-string-surrogate-limit-20261105

  • Reject CIL branch and switch destinations outside the method body without wrapping signed displacements. — claimed: Afloat16, branch bugfix/r3.12-cil-branch-target-ranges-20261111

  • Validate CIL method exception-section kinds, extents and complete clause records before reading exception handlers. — claimed: Afloat16, branch bugfix/r3.12-cil-exception-framing-20261110

  • Preserve distinct nonfinite floating-point constants when hashing normalized CIL instructions instead of collapsing them to JSON null, retaining original single-precision NaN bits before widening. — claimed: Afloat16, branch bugfix/r3.12-cil-nonfinite-hashes-20261110

  • Read Dart version hashes and feature strings only from the bytes belonging to that snapshot. — claimed: Afloat16, branch bugfix/r3.12-dart-snapshot-extent

  • Keep Dart string-pool duplicate tracking within the retained URI and token limits. — claimed: Afloat16, branch bugfix/r3.12-dart-retained-string-budgets-20261110

  • Read a Dart ELF build ID only when its name and descriptor fit the declared note. — claimed: Afloat16, branch bugfix/r3.12-dart-note

  • Locate ELF core mapping names after the complete declared table. — claimed: Afloat16, branch bugfix/r3.12-elf-core-file-name-count-cap-20261105

  • Advance ELF core NT_FILE path cursors by complete original byte strings, even when a displayed path is lossy or capped. — claimed: Afloat16, branch bugfix/r3.12-elf-core-path-cursor-20261109

  • Validate index source map section offsets and stop an unresolved section from borrowing mappings from the preceding section. — claimed: Afloat16, branch bugfix/r3.12-index-map-boundaries

  • Preserve the URL delimiter in source-map roots when resolving a relative source file. — claimed: Afloat16, branch bugfix/r3.12-source-map-url-root

  • Apply a nonempty source-map root prefix before interpreting a source path, including leading slashes and URL-shaped sources, while preserving empty-root compatibility. — claimed: Afloat16, branch bugfix/r3.12-source-root-prefix-20261111

  • Preserve embedded source-map contents when safe recovery paths collide, keeping each source identity and index in the manifest. — claimed: Afloat16, branch bugfix/r3.12-source-recovery-collisions-20261111

  • Bound retained path bytes in source-map recovery manifests before repeating expanded source roots, and report partial recovery. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-source-recovery-path-budget-20261111

  • Bound the source-map parser's retained sources/sourcesContent entries before building millions of small strings from a 64 MiB map. — claimed: NicolasBohn, branch bugfix/r3.12-sourcemap-entry-bounds

  • Bound the number of index-map sections the source-map parser keeps; each section, even an empty external one, still becomes a map of its own. — claimed: NicolasBohn, branch bugfix/r3.12-sourcemap-entry-bounds-review

  • A malformed source map (BadJson, BadVersion, bad mappings or sections) and a stand-alone .map file without a comment pointing at it still leave their sources looking absent in an application graph: say they couldn't be read, so a source comparison doesn't confirm those sources missing. — claimed: NicolasBohn, branch bugfix/r3.12-unreadable-source-maps

  • A source map refused for bad mappings or sections still has a readable sources list: keep those sources with a warning instead of reporting the whole map unreadable. — claimed: NicolasBohn, branch bugfix/r3.12-sources-of-bad-mappings

  • An index-map section that points at an external map (url) adds no sources. The parser counts those unread sections, an application graph records them with a limitation, and a source comparison reports otherwise unmatched sources as uncertain (source_map_external_sections) with partial coverage instead of missing. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-source-map-external-sections

  • Recognize source-map annotations with no space, tabs or repeated spaces after the marker, and reject whitespace inside their URLs. — claimed: Plastic-time, branch bugfix/r3.12-sourcemap-annotation-spacing-1111

  • Keep source map annotation URL suffixes intact. — claimed: Plastic-time, branch rrfz/bugfix-r3.12-annotation-20261110

  • Match HTML script types against complete JavaScript MIME names and the exact module keyword. — claimed: Plastic-time, branch bugfix/r3.12-html-types-1116

  • The JavaScript crate's sourceMappingURL reader is the formats reader, so the standard no-space annotation is found, tabs and repeated spaces are accepted, and a URL containing whitespace is rejected, on the parsed and lexer paths alike. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-js-sourcemap-annotation

  • Keep DOS relocations inside the declared executable header. — claimed: Afloat16, branch bugfix/r3.12-dos-relocation-header-boundary-20261105

  • Report contradictory DOS executable size declarations as header problems. — claimed: Plastic-time, branch rrfz/bugfix-r3.12-mz-header-20261110

  • Keep relocation reads within their declared entry stride and never invent an addend for a truncated RELA. — claimed: Afloat16, branch bugfix/r3.12-sbpf-rela

  • Use the sorted section and segment boundaries during source-map position lookups so repeated symbolication visits only matching positions. — claimed: Afloat16, branch bugfix/r3.12-source-map-lookup-work

  • Bound generated source-map lines across indexed sections, including empty lines that consume no segment budget. — claimed: Afloat16, branch bugfix/r3.12-sourcemap-line-budget-20261110

  • Parsers refuse what they used to half-read: binary plists keep objects inside their object area and reject unrepresentable widths, base64 (plists, source maps) refuses damaged tails and incomplete quanta through one shared decoder, OIDs must be minimally encoded, and string offsets past the address space are left out. Tar keeps old GNU timestamps out of names and PAX sizes across other metadata records; CRC cpio checks member data; source maps decode percent-escaped data URLs and report indexed-map columns in the generated file (#606–#615, by Afloat16)

  • Use the existing bounded raw LZ4 decoder for SquashFS metadata, regular data blocks and tail fragments while retaining Unity's exact decoded-size contract. — claimed: Afloat16, branch feature/r3.12-squashfs-lz4-20261110

  • Report truncated SquashFS metadata blocks and retain previously recovered blocks. — claimed: Afloat16, branch bugfix/r3.12-squashfs-metadata-warning

  • Require each raw or decoded file block to cover exactly its logical range, including the partial final block, so short blocks cannot shift later contents and long blocks cannot be hidden by truncation. — claimed: Afloat16, branch bugfix/r3.12-squashfs-block-output-size-20261110

  • Recover zero-filled cramfs hole blocks at full and partial page boundaries while retaining the existing recovery budgets. — claimed: Afloat16, branch bugfix/r3.12-cramfs-hole-blocks-20261110

  • Charge firmware ZIP outputs before decoding the next member, so byte and file limits bound retained member data. — claimed: Afloat16, branch bugfix/r3.12-firmware-zip-output-budget-20261110

  • Skip inconsistent JFFS2 data nodes and recover files when later valid nodes cover their damaged extents within the final file size. — claimed: Afloat16, branch bugfix/r3.12-jffs2-decoded-node-lengths-20261111

  • Require complete, structurally consistent extent nodes and child depths before returning a file mapping; missing or invalid metadata leaves the file unrecovered instead of replacing unknown ranges with zeros. — claimed: Afloat16, branch bugfix/r3.12-ext4-unavailable-extent-map-20261110

  • Bound filesystem and archive recovery by the recorded image, node, directory and member limits. — claimed: Afloat16, branch bugfix/r3.12-filesystem-archive-bounds

  • Use the version-two cramfs size to bound root, directory and data reads in either byte order while preserving version-one compatibility. — claimed: Afloat16, branch bugfix/r3.12-cramfs-v2-image-size-20261110

  • Mark cramfs recovery incomplete when a file block-pointer table is unavailable. — claimed: Afloat16, branch bugfix/r3.12-cramfs-pointer-completeness-20261111

  • Recover cramfs direct block pointers, including shared uncompressed pages and length-prefixed compressed blocks. — claimed: Afloat16, branch feature/r3.12-cramfs-direct-block-pointers-20261111

  • Recover files and symlinks from cramfs blocks marked uncompressed while preserving mixed compressed runs and recovery limits. — claimed: Afloat16, branch bugfix/r3.12-cramfs-uncompressed-blocks-20261110

  • Follow the third classic ext block-pointer level, preserving the full logical span of sparse subtrees and refusing unavailable mapping blocks. — claimed: Afloat16, branch feature/r3.12-ext-triple-indirect-20261110

  • Preserve ext device, socket and FIFO inode kinds without exposing their block fields as regular-file contents. — claimed: Afloat16, branch bugfix/r3.12-ext-special-inode-kinds-20261111

  • Use ordered visited-node membership in ROMFS, ext, SquashFS and JFFS2 directory walks so wide filesystem trees avoid repeated scans of every previously visited node. — claimed: Afloat16, branch bugfix/r3.12-filesystem-page-visits-20261110

  • Read JFFS2 node, inode and directory integers in the filesystem byte order, including images with erased leading space, and disclose mixed byte orders. — claimed: Afloat16, branch feature/r3.12-jffs2-big-endian-20261110

  • Read ROMFS names at the configured name-length limit. — claimed: Plastic-time, branch rrfz/bugfix-r3.12-romfs-name-20261110

  • Read regular ext4 inline files up to the 60-byte inode payload while preserving byte budgets; unsupported inline continuations and directory layouts remain explicitly unrecovered. — claimed: Afloat16, branch feature/r3.12-ext4-small-inline-files-20261110

  • Read big-endian cramfs inode fields and block pointers so files and links recover like little-endian images. — claimed: Afloat16, branch bugfix/r3.12-cramfs-big-endian-20261110

  • Verify gzip FHCRC and accept informational XFL and OS bytes without inventing reserved-value restrictions. — claimed: Afloat16, branch bugfix/r3.12-gzip-header-crc-20261109 Source dde4aa02 passed formats tests, fmt and clippy and layers, locks, secret scanning, wasm checks and Cargo.lock stability. The unchanged-code baseline reproduces the new regression cases.

  • Detect ZIP members through the archive directory, not signatures inside nested files or comments. — claimed: Plastic-time, branch bugfix/r3.12-zip-member-detection-1113

  • Match firmware ZIP end records to the same archive start so stored nested archives do not shorten the outer region. — claimed: Afloat16, branch bugfix/r3.12-firmware-zip-extent-20261110

  • Exercise signed PAX lengths, empty keys and equals signs inside a path value. — claimed: Afloat16, branch bugfix/r3.12-pax-record-framing

  • Reject invalid local PAX sizes before listing the affected TAR member. — claimed: Plastic-time, branch bugfix/r3.12-tar-local-pax-size-1112

  • Apply global PAX archive attributes to later members, honor local overrides and global deletions, and bound inherited name copies before extraction.

  • Pass recovered filesystem file bytes into firmware extraction without retaining a second full copy. — claimed: Afloat16, branch refactor/r3.12-firmware-owned-payloads-20261110

  • Bound and cancel CRC cpio verification across firmware probes, and reject non-hex checksum fields — claimed: Afloat16, branch bugfix/r3.12-bound-cpio-crc-scans-20261103

  • Distinguish short block-backed ext symlinks from inline symlinks using inode allocation metadata. — claimed: Afloat16, branch bugfix/r3.12-ext-block-backed-short-symlinks-20261111

  • More parsers keep to the bytes they declare and refuse impossible values: signed LEB128 past 64 bits, damaged modified UTF-8 (valid text around it survives), widened Java instructions and invocation operands, DEX register lists and array widths, SQLite text in the database's encoding with cells kept inside their page and overflow cycles stopped, Android string pools, ZIP output limits for stored entries, decoded sizes and ZIP64 fields, ASAR header framing, ROMFS dot roots, cramfs, ext and JFFS2 sizes and holes, PyInstaller modules outside their PYZ, JSON config edits (comments kept, every duplicate key removed), and UDP, TCP, DNS over TCP, pcapng and TLS lengths (#688–#715, by Afloat16)

  • Check ClientHello vector lengths and extension values before reporting the handshake as complete. — claimed: Afloat16, branch bugfix/r3.12-tls-client-hello-values

  • Report SQLite row counts and reads as incomplete when b-tree children are null, cyclic or reused. — claimed: Afloat16, branch bugfix/r3.12-sqlite-tree-completeness-20261110

  • SQLite tables whose quoted name holds a parenthesis ("export (1)") keep their columns, rowid alias and WITHOUT ROWID order: the column list opens at the first parenthesis outside quotes. — claimed: Plastic-time, branch bugfix/r3.12-sqlite-parenthesized-table-names

  • Use ordered page membership for SQLite row and count walks so a wide b-tree does not repeatedly scan every previously visited page. — claimed: Afloat16, branch bugfix/r3.12-sqlite-page-visits-20261110

  • A comment after a SQLite column list that holds ) (... WITHOUT ROWID /* ) */) still hides the WITHOUT ROWID marker: find the list's closing parenthesis by tokens, not rfind(')'). — claimed: Chr0mium24, branch bugfix/r3.12-sqlite-options-after-comment

  • Keep Android binary XML events within their document, chunk and attribute bounds. — claimed: Afloat16, branch bugfix/r3.12-android-xml-events

  • Reject Java Long and Double constants without their reserved constant-pool slot, including the maximum u16 pool boundary. — claimed: Afloat16, branch bugfix/r3.12-java-wide-pool-slot-20261109 Source e0dbc3a5b60b; full Linux gate passed.

  • Borrow Java attribute names from the constant pool so repeated references cannot multiply temporary parser memory. — claimed: Afloat16, branch bugfix/r3.12-java-attribute-name-borrowing-20261110

  • Unfold JAR manifest attributes as bytes before UTF-8 decoding, preserve value whitespace, and recognize every specified line ending so entry digests use the original filenames. — claimed: Afloat16, branch bugfix/r3.12-jar-attribute-bytes-20261110

  • Merge digest attributes across repeated JAR manifest sections for the same entry, honoring later attribute values while retaining absent earlier values and checking each file once. — claimed: Afloat16, branch bugfix/r3.12-jar-repeated-entry-sections-20261110

  • Connect class literals, method types, method handles and dynamic bootstrap arguments to their actual load or call instruction, with a shared linkage-work limit. — claimed: Afloat16, branch feature/r3.12-jvm-bootstrap-references

  • Find the method parameter terminator outside object type names in JVM descriptors. — claimed: Afloat16, branch bugfix/r3.12-java-descriptor-boundary

  • Validate complete JVM switch payloads before reserving pairs, and reject duplicate or descending lookup keys instead of emitting invalid control-flow facts. — claimed: Afloat16, branch bugfix/r3.12-java-switch-tables-20261110

  • Stop JVM bytecode decoding on invalid primitive array type codes or zero-dimensional multidimensional allocations while retaining valid allocations. — claimed: Afloat16, branch bugfix/r3.12-java-array-operands-20261110

  • Bound aggregate DEX payload copies while attaching shared switch and array data to decoded instructions. — claimed: Afloat16, branch bugfix/r3.12-dex-payload-memory-20261110

  • Stop capped DEX member and handler lists from interpreting unread records as the following field or a catch-all. — claimed: Afloat16, branch bugfix/r3.12-dex-capped-table-alignment-20261109 Source 149f060b9959; format checks and compatibility gates passed.

  • Address the combined parser review: preserve partial Java reference coverage, enforce DEX signed 32-bit LEB128 limits, recover legacy AAPT UTF-8 lengths within their pool with a shared recovery probe limit, and distinguish missing ext indirect metadata from sparse holes. — claimed: Afloat16, branch bugfix/r3.12-parser-review-followups-main-20261104 Shared-budget revision 29c48e03 passed full Linux CI. The merge with main 41d2175e awaits a new CI run.

  • Recognize HTTP extension method tokens in packet captures, including WebDAV requests. — claimed: Plastic-time, branch bugfix/r3.12-http-extension-methods-1112

  • Expose SVCB/HTTPS target and service parameters, SSHFP/TLSA authentication data, and NAPTR/URI/DNAME discovery values through bounded DNS answer decoding. — claimed: Afloat16, branch feature/r3.12-dns-service-bindings

  • Keep DNS answer names inside their declared RDATA, including compressed names and MX exchanges, and report malformed records as incomplete instead of inventing answer values, while preserving legal empty DNS UPDATE RRset prerequisites — claimed: Afloat16, branch bugfix/r3.12-dns-rdata-names-review-20261104

  • Keep address and service records attached to their owner and validate the record's wire fields. — claimed: Afloat16, branch bugfix/r3.12-dns-record-evidence

  • Resolve keyed archive references consistently in nested objects and object rows. — claimed: Afloat16, branch bugfix/r3.12-keyed-archive-references

  • Recognize UTF-16 XML keyed-archive NIBs after either byte-order mark so their objects, classes and references reach the existing plist decoder. — claimed: Afloat16, branch bugfix/r3.12-nib-utf16-detection-20261111

  • Distinguish the keyed archive's null sentinel at object zero from ordinary stored strings containing the same text. — claimed: Afloat16, branch bugfix/r3.12-keyed-archive-null-string-20261110

  • Reject NIB variable-length integers whose tenth byte exceeds the unsigned 64-bit range, instead of letting truncated high bits become an unrelated length or index. — claimed: Afloat16, branch bugfix/r3.12-nib-integer-overflow-20261110

  • Accept binary plist arrays and sets that name one stored value many times; repeated data and text spend a byte budget, and containers past the node cap are refused before reserving space. — claimed: Plastic-time, branch bugfix/r3.12-plist-shared-array-values

  • Bound NIB object copies by their retained data, key and class bytes so shared value ranges cannot amplify a small archive beyond the byte budget. — claimed: Afloat16, branch bugfix/r3.12-nib-retained-bytes-20261110

  • Use the complete SQLite type spelling to decide whether a quoted primary key aliases the rowid. — claimed: Plastic-time, branch bugfix/r3.12-sqlite-type-boundaries

  • Read a quoted SQLite column type as its type name, including an INTEGER rowid alias. — claimed: Plastic-time, branch bugfix/r3.12-sqlite-quoted-types

  • Separate SQLite column names from their types at carriage returns and form feeds. — claimed: Plastic-time, branch bugfix/r3.12-sqlite-column-whitespace-1112

  • SQLite column names read whole when quoted: a doubled quote inside one stands for a quote, and single-quoted names (which SQLite accepts) lose their quotes. Keys that name such columns find them, so rowid aliases and WITHOUT ROWID order hold. — claimed: Plastic-time, branch bugfix/r3.12-sqlite-quoted-column-names

  • Removing the last member of a JSONC object takes the comma before it too, so a config that is plain JSON (Claude Code's, Claude Desktop's) stays plain JSON after an uninstall; an object written with trailing commas keeps them, new members included. — claimed: Plastic-time, branch bugfix/r3.12-jsonc-no-new-trailing-comma

  • Preserve CR-only JSONC line endings and member indentation when inserting or removing settings. — claimed: Plastic-time, branch bugfix/r3.12-jsonc-cr-lines-1113

  • Preserve CRLF when creating the first JSONC object in a whitespace-only or comment-only file. — claimed: Plastic-time, branch bugfix/r3.12-jsonc-first-object-eol-1111

  • SQLite rows keep each value on its own column when a table has VIRTUAL generated columns: SQLite computes those when a row is read and stores no field for them, so they show as not read (with a note) instead of shifting later values onto the wrong columns, in rowid and WITHOUT ROWID tables alike. — claimed: Chr0mium24, branch bugfix/r3.12-sqlite-virtual-generated-columns

  • A SQLite CREATE TABLE longer than 4 KiB is read whole. The schema walk cut it at the 4 KiB text cap, so the columns after the cut weren't listed, a rowid alias read NULL, a WITHOUT ROWID table showed no rows, and the defaults of columns added later weren't filled. — claimed: Chr0mium24, branch bugfix/r3.12-sqlite-long-create-table

  • Comments in a SQLite CREATE TABLE stay out of its columns: a comma or quote in a comment doesn't split or swallow definitions, a comment before or right after a name isn't part of it, and a comment inside a definition isn't part of its type, so INTEGER PRIMARY KEY still marks the rowid alias. — claimed: Plastic-time, branch bugfix/r3.12-sqlite-create-table-comments

  • Keep the comma inside a SQLite type's parentheses: DECIMAL(10, 2) is listed as DECIMAL(10 2). The type keeps the comma inside its parentheses, and the schema overview shows it unchanged. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-sqlite-type-comma

  • Keep shortened IL2CPP literals within their UTF-8 output budget without inventing a partial character. — claimed: Afloat16, branch bugfix/r3.12-il2cpp-literal-byte-budget-20261109 Source d81b7d570546; format checks and compatibility gates passed.

  • Check that IL2CPP method ranges agree with each method's declaring type before reporting names and ownership. — claimed: Afloat16, branch bugfix/r3.12-il2cpp-method-owners

  • Asset catalogs name each rendition key attribute as CoreUI reads it. When the deployment target predates an attribute (display gamut before macOS 10.12), actool writes a stand-in (the placeholder attribute) in KEYFORMAT and the real one in KEYFORMATWORKAROUND, which isn't read: the key format lists placeholder and a Display P3 variant's key shows placeholder: 1 instead of display-gamut: 1. — claimed: Chr0mium24, branch bugfix/r3.12-car-key-format-workaround

  • Swift optionals of function types keep their parentheses. The demangler printed the type inside ? or ! as is, so an optional closure read as a closure that returns an optional: WeatherKit.WDSEndpoint.portOverrideProvider read () -> Swift.Int? where swift-demangle prints (() -> Swift.Int)?. An optional protocol composition read main.P & main.Q? for (main.P & main.Q)?. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-optional-functions

  • Swift special types decode. A type written with X and a kind letter (a non-escaping closure XE, an AnyObject or class-and-protocol existential, a protocol metatype, a @convention or @autoclosure function, Self, a weak or unowned type) was reported invalid ("unsupported special type"): _$s13IDEConsoleKit0A5ThemeC22infoTypeThumbnailImageSo7NSImageCvgAFyXEfU_ is closure #1 () -> __C.NSImage in IDEConsoleKit.IDEConsoleTheme.infoTypeThumbnailImage.getter : __C.NSImage. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-special-types

  • Python 3.5.0, 3.5.1 and 3.5.2 wrote magic number 3350 into their .pyc files, and 3.5.3 moved to 3351. A file with 3350 is reported as a 3.5 pre-release (is_release: false), with a note that its code layout may differ. Count 3350 and 3351 as releases of 3.5. — claimed: Chr0mium24, branch bugfix/r3.12-pyc-3-5-release-magic

  • Report undefined Python bytecode header flag bits while preserving the code inventory and the defined invalidation mode. — claimed: Afloat16, branch bugfix/r3.12-pyc-flag-validation-20261110

  • Wrap extracted PYZ code with the header length required by its Python magic number; preserve the marshal payload and apply the same version-specific size to extraction limits. — claimed: Afloat16, branch bugfix/r3.12-pyz-version-headers-20261110

  • Swift addressors decode. After a (mutable) or l, an addressor has a letter for its kind (u for unsafe), but the demangler read only the a and took the kind letter for another operator, so every addressor was reported invalid: _$s5Books19PAGETURNLOG_ENABLEDSbvau is Books.PAGETURNLOG_ENABLED.unsafeMutableAddressor : Swift.Bool. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-addressors

  • Swift outlined variables decode. Tv with an optional index names the outlined variable of the symbol before it (a function, a protocol witness, or a global), but the demangler reported every one as an unsupported thunk, so _$s013CompilerSwiftA21PluginMessageHandling0C7FeatureO8rawValueACSgSS_tcfCTv_ is outlined variable #0 of CompilerSwiftCompilerPluginMessageHandling.PluginFeature.init(rawValue: Swift.String) -> CompilerSwiftCompilerPluginMessageHandling.PluginFeature?. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-outlined-variable

  • Swift async thunks decode. Tu is the async function pointer of the function before it, and TY and TQ with an index are its suspend and await resume functions, but the demangler reported all three as an unsupported thunk, so _$s10XCTestCore5QueueC11synchronizeyyYaFTY0_ is (1) suspend resume partial function for XCTestCore.Queue.synchronize() async -> (). — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-thunks

  • Swift subscript accessors print the accessor after subscript and the type after a colon, as swift-demangle does. The demangler put the accessor last, so _$s10Foundation4DataV06InlineB0Vys5UInt8VSiciM read Foundation.Data.InlineData.subscript(Swift.Int) -> Swift.UInt8.modify where swift-demangle prints Foundation.Data.InlineData.subscript.modify : (Swift.Int) -> Swift.UInt8. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-subscript-accessors

  • Swift key path accessors decode. The getter and setter of a key path (TK and Tk) after a property descriptor and the type the path starts from were reported as an unsupported thunk, so _$s10DVTIconKit18GeneratedIconImageC14drawingContextAA0cd7DrawingG0CSgvpACTk is key path setter for DVTIconKit.GeneratedIconImage.drawingContext : DVTIconKit.GeneratedIconDrawingContext? : DVTIconKit.GeneratedIconImage in swift-demangle. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-keypath-thunks

  • Swift reflection, metadata-cache and Objective-C class symbols decode. The demangler stopped at M followed by K, D, F, A, P, B, U, t or s and reported the symbol invalid ("unsupported metadata kind"); about 206,000 symbols in macOS 15.7 and Xcode 26 system binaries were affected. _$s7Testing3TagVSeAAMcMK is metadata instantiation cache for protocol conformance descriptor for Testing.Tag : Swift.Decodable in Testing. — claimed: Chr0mium24, branch bugfix/r3.12-swift-demangle-metadata-kinds

  • Name the glyph weight (26) and glyph size (27) key attributes of asset catalogs, which hold SF Symbol style images; they're listed as attribute-26 and attribute-27 today. — claimed: Chr0mium24, branch bugfix/r3.12-car-glyph-attributes

  • Reject malformed ASAR directory, link and file nodes instead of listing fabricated empty files. — claimed: Afloat16, branch bugfix/r3.12-asar-node-shapes-20261109 Source 351002ea passed formats tests, fmt and clippy and layers, locks, secret scanning, wasm checks and Cargo.lock stability. The unchanged-code baseline reproduces the new regression cases.

  • Reject duplicate GGUF metadata keys and tensor names before they overwrite metadata or inflate parameter totals. — claimed: Afloat16, branch bugfix/r3.12-gguf-duplicate-identities-20261111

  • Calculate GGUF quantized tensor byte ranges from their block layout and diagnose invalid row widths. — claimed: Afloat16, branch bugfix/r3.12-gguf-quantized-extents-20261110

  • Require complete sBPF symbol records and their declared string-table links before reporting symbol identities. — claimed: Afloat16, branch bugfix/r3.12-sbpf-symbol-framing-20261110

  • Keep the sBPF symbol and relocation output limits global across repeated ELF tables. — claimed: Afloat16, branch bugfix/r3.12-sbpf-table-records

  • Bound retained PE import/export name and forwarder bytes across repeated references; preserve bounded prefixes and report byte-budget truncation.

  • Refuse padded and unterminated subidentifiers and cut-off arc lists in code-signature requirement OIDs instead of showing a different OID, and split a long first subidentifier. — claimed: NicolasBohn, branch bugfix/r3.12-codesign-oid-subidentifiers

  • Require call syntax for tRPC client hooks in JavaScript lexer fallback facts. — claimed: Plastic-time, branch bugfix/r3.12-trpc-calls-1116

  • Keep side-effect imports in JavaScript lexer fallback facts. — claimed: Plastic-time, branch bugfix/r3.12-js-side-effect-imports-1115

  • Bound the JavaScript lexer fallback's endpoints (fetch calls and URL literals) and exposed lists to the same MAX_LIST cap as the syntax-tree pass, and record endpoints in truncated when a fetch call is dropped, so a huge or unparsable bundle can't grow the lists without limit. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-js-fallback-list-bounds

  • A source-to-bundle comparison with more scripts than max_rows now says how many it examined (scripts_examined), keeps the unclaimed-unit total complete, marks rows_capped, and adds a partial-coverage limitation, so a capped window's counts are never read as totals. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-source-bundle-compare-totals

  • app_graphs_compare and js_exports_compare cut added, removed, bundle-module, vendor and changed-module function/export rows at 2,000 without saying so, and the app comparison's summary reports the cut lengths as totals; report complete counts, mark rows_capped, and add partial coverage. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-compare-row-totals

  • app_graphs_compare and js_exports_compare drop the loaded graph's analysis limitations (a malformed source map, a truncated inventory, dynamic IPC names), so a comparison reads as complete when a side was partial; carry each side's limitations into the comparison evidence with left/right labels. — claimed: AlexJoaquimPereira, branch bugfix/r3.12-compare-graph-limitations

  • Take target file signatures from the shared craft-sniff table. The app still breaks ties that need its own parsers (managed PE, Java class versus a universal Mach-O, APK by name or manifest, ASAR, text). Fat Mach-O files that start 0xCAFEBABF are now recognized. — claimed: FredApps, branch refactor/target-kind-craft-sniff

R3.13 Web inspection correctness

Owner: open · Status: in progress

  • Decode character references in HTML page titles before returning their displayed text. — claimed: Plastic-time, branch rrfz/bugfix-r3.13-html-title-entities
  • Report received HAR response body sizes separately from decoded content sizes and identify entries whose sizes are unknown. — claimed: Afloat16, branch bugfix/r3.13-har-transfer-size
  • Read HAR captures with the permitted UTF-8 byte order mark. — claimed: Afloat16, branch bugfix/r3.13-har-bom-bounds-20261107
  • Require complete HTML start tags before recording scripts, preloads and page metadata. — claimed: Plastic-time, branch bugfix/r3.13-html-incomplete-tags
  • Resume scanning after an HTML comment that a browser closes with --!> or an abrupt empty end. — claimed: Plastic-time, branch bugfix/r3.13-html-comment-endings
  • Skip bogus HTML comments opened by unknown declarations and malformed end tags before listing scripts or preloads. — claimed: Plastic-time, branch bugfix/r3.13-html-bogus-comments
  • The HTML scanner finds script preloads in rel token lists, decodes asset URLs once (numeric and named references), keeps script and title text through lookalike closing tags, and skips raw-text elements, markup inside attribute values and custom tag names such as <script:widget> (#591–#596, by Afloat16)
  • HAR reports redact credentials whose query names are percent-encoded (and the stand-in backend matches older recordings the same way), redact declared secrets in every kept field (names, methods, hosts, versions, timings), count body bytes before text conversion, and template only real calendar dates in API paths (#602–#605, by Afloat16)
  • Preserve empty query pairs when matching older stand-in recordings, without treating raw query separators as pairs — claimed: Afloat16, branch bugfix/r3.13-preserve-empty-query-pairs-20261103
  • A walkthrough's keep_bodies keeps response bodies only for the first 200 requests and says nothing about the rest (#1811; web_page_show's outlines were fixed in #1845): count the kept bodies and say how many were left out and why. — claimed: htcom-code, branch bugfix/r3.13-body-outline-cap

M4 Release

R4.1 A macOS CLI that opens

Owner: maintainers · Status: todo

  • decision A standalone macOS CLI that can carry its ticket, such as a stapled .pkg; that needs a Developer ID Installer certificate in the release secrets

R4.2 Managed tier

Owner: maintainers · Status: todo

  • decision Managed tier: account backend, usage metering (Claude API under our own key only)
  • Before the managed tier or any paid offering: confirm the Claude Code orchestration question with Anthropic (claude-code.md)

Parked

Looked at and not committed to; none of this is open work. Say in the issue if you want one.

  • From the tool survey: firmware image kinds (ESP32, U-Boot, device trees, Intel HEX), Erlang BEAM, GPU code objects, kernel modules, GameCube DOL, Mac crash reports, SAZ and qlog captures, protobuf bodies, R8 mappings, JNI links, MCP over HTTP
  • Ideas from issues: resume by itself after a usage limit (#476, the one tied to a real problem), a source folder as a target (#480), actions after a build ends (#495), WebAssembly disassembly (#175), an HTTPS intercepting proxy (#164), VM and Windows Sandbox detonation (#144, #163), an adapter for another disassembler (#147)
  • More providers and agents (#14, #63, #97, #162, #170, #186, #206, #470): R2.2's table comes first, so each new one is a small addition rather than another half-wired agent