The DER reader accepted redundant length encodings (a long form under 128, or a leading zero length byte) without saying so. Each element now records whether its length is minimal; certificate, CMS, Authenticode and certificate-chain views keep everything they read and add one note that an element is not strict DER, and signatures over those bytes are checked as before.
A modern .pyc header's flag word defines only its two low bits, and other bits were read without comment. Undefined bits are now reported as a header warning, while the invalidation mode, source fields and code are read from the defined bits as before, so a file from a future Python release that adds a flag stays readable. Legacy headers, which have a timestamp in that position, are unaffected.
POSIX makes everything after = in a ps -o argument the header, so FreeBSD's ps read pid=,ppid= as a single pid column headed ",ppid=". The process tree then had no parents, a runner's child window was never found there, and snapshot_finds_an_owned_child_on_this_host failed on every FreeBSD run since it was added.
A text file outside a build (a script, Markdown, settings, a log) could only be searched byte by byte: the native tools refused it and build_file_read works only in a build. The new core_text_read (Deep mode) reads one as text, a page of lines at a time: UTF-8, or UTF-16 with a byte-order mark, invalid bytes shown as U+FFFD and counted, very long lines cut and named, and a file with zero bytes near its start refused as binary. Nothing is run.
Fixes#539
When a run's time limit or a cancel fired, the program was stopped through a call that then waited for it with no limit (and on Windows for a job message the system doesn't promise to send), so a program that outlived its stop held the run past every deadline, Pause included. The program and whatever it started are still stopped, but the run now waits at most 5 seconds for them to be gone, then logs a warning and carries on.
Refs #1438
A job of unknown length drew its progress bar with egui's animated ProgressBar, which asks for a frame every frame, so the window redrew at the display's rate for as long as the job ran, even in the background (as the spinner did before #579). The jobs list now uses our own progress bar: filled to a known fraction with no frames asked for, or sweeping at the animation's pace (once a second in the background) when the fraction isn't known.
A coding agent's turn ended only when its process exited, so when Claude Code printed its final result and then kept running (a shutdown hook, logging), a finished build stayed on "Building" or "Pausing" for as long as the idle limit allowed, hours for a builder turn. The turn now ends a short grace (5 s) after the stream's own result, without being reported as stopped or failed; a deadline hit before any result still stops the turn as before.
When a tar listing stopped at a GNU long name or link it refused (too long, or cut short), the engine installer still wrote the files listed before it and reported success, which could replace a working engine with part of one. The listing now says why it stopped, and the installer refuses such an archive before writing anything, with a plain-English error. An archive missing only its end marker still installs.
A private browser or driven Electron app was stopped only when its process handle was dropped, which a crash, a kill, a force quit or an MCP server stopped by its client never does, so the browser was left running and its profile stayed in the temp folder. On Unix a small watcher now stops the browser's process group and deletes its `craft-*` profile once the app is gone, and ends by itself as soon as the browser does. Windows children are held by a Job Object; whether one outlives an abrupt exit is left as a roadmap line.
Refs #810
Claude Code reports the whole session's running cost on every turn, and a build resumes the same session each turn, so adding each turn's figure made a long build's estimate grow far past what it cost. The last total of each session is now kept in the config folder, and a resumed turn counts only what it added; a new session's first turn counts its total, and a resumed or forked session with nothing kept counts nothing rather than an inflated figure.
Refs #1187
A round of the builder's own checks now has a 30-minute limit, Pause is looked at before each check and stops a running build step, and every check is said as its own step ("Checking its work (2 of 4): …"), so a check that stalls names itself instead of holding the build for hours. The limit reaches the checks' own program runs, a step is refused before it starts once the round is past its limit or paused, a check that can't run is reported as not run with its reason, and a caught bug leaves the project at "Needs you" instead of "Checking". Real check rounds keep no limit. Addresses the stalls reported in #981 and #1198; a live Windows run is still to do.
OpenCode Go backends that weren't added through Settings (from an older version, or found from an environment key) had no saved session id, so each run sent a new one. The app and the CLI now save one at startup for any OpenCode Go backend that has none, before any session lock is taken, and keep it across restarts. An OpenCode error that comes without a message now shows its name, HTTP status and the start of the response body instead of "no error details".
Refs #1225
With no `max_tokens`, OpenRouter holds a model's whole output limit against the key's credit, so a key that still had money was refused with 402 "can only afford N" before the model ran. On that refusal the request is now sent once more with the answer capped at what the key can afford, and the cap is kept for the conversation. A cap under 4,096 tokens, a repeat refusal and other providers stay a billing error as before.
Fixes#1213
Ollama fits a prompt that is too long for its context by dropping the oldest messages; once the person's request is among them, a Qwen 3.8 model refuses what's left with 500 "no user query found in messages". That refusal is now reported as a context overflow, saying how to give the model a larger context or send less, and it is no longer retried, since the same request would be refused again. Other server errors are still retried.
Fixes#1245
OpenCode model discovery, which also runs before every OpenCode turn, now uses --standalone unless the person chose a --server. Plain opencode api started OpenCode's shared service, which resumes every unfinished session on the computer, in any project, and carries on with its commands.
Part of #1115.
Five STATUS rows that merges had kept twice, with different updates in each copy, are now one row each, the most complete copy (OpenCode models and thinking now says once that SSH runs get the tools), and opencode.md loses its older doubled paragraph. A pitfall says how to merge a conflicting table row.
A reply in Qwen's <tool_call> markup that a server handed back as text is read as the calls it makes, with arguments typed from the tool's schema; a call to a tool the model wasn't offered gets a plain answer naming its tools, instead of ending the turn.
A builder in the API loop, which can't open pictures, is no longer sent to look at screenshots: every brief and check that pointed it at a picture tells it what to use instead (capture.json's layouts, the UI tree, or asking the person). Agent apps keep the picture wording.
Git's own author variables are no longer taken for secrets while the hook runs, so a claim that names its author passes; other AUTH-named variables are still checked, and a Unix home folder is listed once.
The two start guard tests share the process-wide held mark, so one's opened() could drop the other's lock mid-test and fail main now and then. They now take a shared test lock.
A note sent while a build runs now carries the latest Just ask answers its mode may show, including through Claude Code's note hook, and they count as told only once the builder has them. A Clean room run never gets Deep answers.
`runtime_ui_read` with `screenshot: true` returns one on-screen window of the pid as a PNG on macOS 14+, restricted by `window_id` when given. Missing permission, a missing window or a failed capture is a limitation with partial coverage, and no permission dialog opens.
Fixes#248
The preamble and task go on stdin instead of the command line, so a multi-line or long task no longer fails in the Windows desktop app's batch launcher.
The test that a stopped launch takes its descendants with it uses a parent and child started without a shell, so it runs the same on Unix and Windows, and a failed start still stops what it launched.
An Approvals menu beside the builder's-questions switch offers Default, Allow, Ask or Refuse for each kind of action, and `/approvals <kind> <choice>` does the same from the prompt box without sending anything to the builder.
The check for the project's records reads a command with quotes and escapes taken out, as the shell joins them, and any mention of `CRAFT_ANYTHING_AGENT` counts as a system change. docs/claude-code.md says these raise the bar and aren't a wall.
A project can allow, ask about or refuse each kind of builder action (web, commands, outside files, publishing, deletes outside the project, system changes, other tools); a kind it hasn't chosen follows Auto mode as before. `project_approvals_set` sets them, and an agent can only make a kind stricter, including through our CLI from its own shell. "Always" on a push, delete or system card allows that whole kind, and pushes are read word by word in every shell form.
New Python, Go and .NET projects build and launch through one catalogue shared by the build check, the screenshot and Play; a window is found through the program's descendants, then by name.
Pasted pictures get a counter in their name; atomic-write temp files are created exclusively and private from the start; stand-in replay samples keep differing response details.
Program starts and console opens share one lock, and a timeout ends the whole process tree first.\n\nCo-authored-by: SoldierPePe <ahmed.m.alnajjar1@gmail.com>
On Windows with nvm and no version in use, npm shims such as `claude.cmd` find Node in nvm's version folders; if it's still missing the error names `node` and `nvm use`.
Loopback test fixtures read complete request headers, negative probes use port zero, and the stand-in's shutdown test checks its answerer's lifetime rather than a port another test may reuse.
A run with an explicit OpenCode model or thinking choice used to be refused. The adapter now writes our MCP server to a temp config file named by OPENCODE_CONFIG and runs `opencode run --standalone`, and reads OpenCode V2's tool call ids so tool calls reach the feed. Runs over SSH still have no Craft Anything tools.
A website stand-in charges its per-body and total byte limits on the text it actually keeps after redaction, and drops repeated samples before charging them, so polling no longer uses up the budget. A cached 304 whose body was left out stays unavailable (R3.11).
An MCP plugin whose `tools/list` answer is one JSON-RPC line over 1 MiB never loaded: the process reader cut long lines into fragments that never parsed, so the plugin timed out. The partial-line cap is now the 16 MiB the MCP client already allows, with a regression test for a multi-megabyte line.
A program whose file is busy (just written, still open for writing in a child another thread is starting, so Linux says Text file busy) is tried again for about 0.3 s before its start fails. Fixes flaky test failures and a start just after an agent's CLI updates.
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 (kept in craft.json as not_working_checks), and with app/ empty the builder is asked to write the project and show its build's output instead of fixing a build that doesn't exist (refs #736).
Supersedes #776.
Co-authored-by: Fábio Soares <fabiodev1123@gmail.com>
The keep-awake helper ends by itself once the app it serves is gone (Windows checks the process every 30 s, macOS uses caffeinate -w), so closing the window during a build no longer keeps the computer awake.
The window opens on Macs whose Metal compiler rejects the shader language version wgpu takes from the macOS version: a patched wgpu-hal 30.0.1 (identical to the released crate except for this probe) compiles a tiny kernel with each version up to the OS's and keeps the first one accepted. A lost device on a Mac no longer points at the Windows-only software renderer setting (fixes#742).
Ghidra is found where people unzip it (Program Files, the home folder, Downloads, Desktop, /Applications, Homebrew's cask and the usual Linux folders), by folder name a set number of levels down, and the guided install uses a copy already on the computer instead of downloading it again (fixes#740).
Leaving "Work on my project" forgets a folder it would edit in place, and a folder dialog that ends after leaving that door is dropped, so a version made on Make can't edit the person's own project. An address in that door's request stays part of the request instead of becoming an original to recreate. Follow-up to #686's review.