The dashboard stamps its liveness every 5 s; job processes expire leases only while
it serves, and for missing heartbeats only after it served the whole idle window.
After a gap every lease's heartbeat, claim and run start move forward by the gap.
A token's drops count at most once in 10 minutes.
The time limit counts from the first running heartbeat; preparing may last 15 min.
A claim repeated with its request id within 5 minutes gets the same reply (kept in
the dashboard's memory; after a restart the orphaned lease goes back unused).
Ended and requeued leases keep their secret hash, so every lease route answers 410.
ok:false requeues with the next attempt and skips that login; a repeated accepted
submit is accepted again. Cool-off is per login, counts drops and 5 failures in a
row within 30 min, and is cleared by login.
- Lanes keep the order they first arrived in; a lane the server stops sending stays as ended until the reader returns to the tab. Quiet steps are left out before the queued LIMIT.
- Open integration batches are lanes, with the build of their combined tree.
- The latest five finished Now feed events sit under the board.
- Account labels on model calls are for staff only, and a lone call's progress line is blanked when it names a hidden repo, on the board and in the Now feed.
- The merge queue row shows its wait label, conflict note and operator approval as text.
- The live query and the item status list follow their topics; the board does not poll.
- Workable tables switch to a wider layout only with room to spare, so a page scrollbar cannot flip them back and forth.
- Security findings get their own icon; item icons name their kind and state without explanations; the board's reading tip is gone.
- The dashboard spec describes the Live board, the feed list under it and the item peek.
The weighted fast track and add-to-queue change itself keeps main's form. On top of it:
- Queue items: a button on the Merge queue tab opens a dialog that takes PRs and issues, one per line, with a weight, and answers each line; refused lines stay in the box.
- An issue line fast-tracks the issue. When exactly one open PR closes it, that PR is queued too; when several do, the answer lists them and none is queued.
- Every fast-track mark Queue items makes keeps a heavier weight already there.
- Each merge queue row says how the PR came in, by whom (staff only) and when, in an Added column.
- The server applies x-autocraft-scope only on routes marked scoped (decisions inbox and needs, items, merge queue and next
steps, activity, Now and the project home), and never on writes. Item, decision, batch, job, agent, finding, search,
settings, /api/repos, /api/me and /api/progress show every repo the user may see.
- A list page's project is its URL's repo (or /projects/<repo>), else the pick; requests carry it and cache keys include it.
A list page with a pick and no repo in its URL puts the pick there; a link with ?repo=X leaves the pick alone.
- /api/decisions/open-counts gives the user's inbox count per repo; the picker and the tiles show it.
- / opens the picked project only on first load; the logo and the Projects link show the tiles; a tile picks its project.
- The phone picker sits in the navigation drawer.
An issue line fast-tracks the issue with the weight, so Fix works it first, and queues each open PR that closes it (GitHub's closing references), saying which were queued or that none closes it yet. The dialog is wider (up to 760px) with room for twelve lines and says what queuing a PR and an issue does; Dialog gains size="lg" for it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The add box at the top of the Merge queue becomes a Queue items button. It opens a dialog that takes PRs one per line (a GitHub URL, repo#n or owner/repo#n) and a weight, adds each in turn and answers per line; refused lines stay in the box to fix.
Merge queue rows carry how the PR came in, from data the database already keeps: Queued (an operator's Queue items, from its pre-approval row), Pre-approved (a pre-approval with a fix-up), Merge answer (decisions.answered_by, or a Merge answer carried to a new head), or Take review. The login shows to staff only. The table's Added column shows it with the time, in place of Updated, and the weight moves into the row's last cell beside its menu.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fast-tracked items have a weight from 1 to 10 (default 5): heavier items
go first in every stage that orders by the work rank, and equal weights
keep the order added. The weight is a new fast_track column; the scores
(fastScoreSql, workScores) put it above the order added, so every planner
and the merge order follow it without other changes. The CLI takes
`fasttrack add <item> --weight N` and `fasttrack weight <item> N`, and
POST /api/fast-track/weight sets it (audited as fasttrack-weight).
"Add to merge queue" (POST /api/queue/add {item, weight}, operators)
starts a pre-approved merge at the PR's current head with no fix-up, a
row in `review` as a carried Merge answer uses, so the PR merges only
after a sure take review there, a clean pre-merge review and green CI; a
take-fix review, a defect or a red head asks the Merge? question. It also
fast-tracks the PR with the weight.
The Merge queue tab's add box now adds to the merge queue with a weight,
the fast-track marker shows the weight, and operators change it on the
board's cards.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fast-tracked items have a weight from 1 to 10 (default 5): heavier items
go first in every stage that orders by the work rank, and equal weights
keep the order added. The weight is a new fast_track column; the scores
(fastScoreSql, workScores) put it above the order added, so every planner
and the merge order follow it without other changes. The CLI takes
`fasttrack add <item> --weight N` and `fasttrack weight <item> N`, and
POST /api/fast-track/weight sets it (audited as fasttrack-weight).
"Add to merge queue" (POST /api/queue/add {item, weight}, operators)
starts a pre-approved merge at the PR's current head with no fix-up, a
row in `review` as a carried Merge answer uses, so the PR merges only
after a sure take review there, a clean pre-merge review and green CI; a
take-fix review, a defect or a red head asks the Merge? question. It also
fast-tracks the PR with the weight.
The Merge queue tab's add box now adds to the merge queue with a weight,
the fast-track marker shows the weight, and operators change it on the
board's cards.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- the pre-merge step takes a slot of the rolling hour before each re-read and reads the core rate once per pass; below the reserve the re-reads pause until the reset
- the plan waits for the pace (an external wait in the registry) and wakes when a slot frees
- the workflows listing serves every PR of a repo and default branch for ten minutes
- the sync's reserve is one constant both use
- while the extras are pending the facts carry the reasons found without them plus a waiting reason; the Merge? question waits
- every open PR with a pending diff gets a re-read by the pre-merge step, at once and then an hour after each incomplete read; the step records the live base as the review does
- only these re-reads count towards the three tries, and other cached-diff readers never fetch again for the extras
- workflow defaults come only from defaults: run: working-directory; a step's expression working directory matches named files as path suffixes
- the facts memo keys on file times and sizes instead of reading the extras each pass
A confidential environment is ready from its worker's first call after the flag, or the environment, was turned on (confidential_since, confidential_online_at); label, slot and key changes do not move the 4-hour build limit of private tasks. Until one is ready the fix job ends before the model call, the finding notes the wait and advanceSecurityFixes continues it once one is ready, and the confidential-env health check warns while security fixes or security build tasks wait; build-queue leaves those tasks to it.
- TOML tables have no prototype; __proto__, constructor and prototype keys make a manifest unreadable, so it is held
- poetry checksums come from files entries only and [metadata] is skipped; a checksum change the hunks cannot attribute is decided from the lockfile's head text
- run-step tokens match exact files, joined with the step's or the default working directory; only local actions match by directory; xtask sources count when a step runs xtask
- package.json scripts changes need a person; build dependencies hold on an added, removed or renamed entry or a non-registry source, not on a registry bump; a byte-order mark is stripped
- workflows are read at the default branch's head, cached per blob, and only when a changed path could be run by name
- a cached diff that lacks what the rule reads beside it, or whose read failed, waits: the pre-merge step reads it again without a model call, at most three times, then the rule judges with what it has
- fetchScript() builds the fetch prologue (image toolchain, repository cargo config removed, env -i) for any fixed command list; rustPlan and the nightly fetch deps, fetch advisories and build-std fetch steps use it.
- A dangling symlink is kept when its target, walked component by component from the link's directory, never leaves the tree and passes no existing symlink; otherwise it is removed.
- The alert pass reads a maintainer's repo permission from GitHub (the normal read path) when the
cached one is over a day old, at most once a day per login and repo. A failed read keeps the
older row; only an answer without write access ends the alerts. Turning alerts on takes a
write permission of any age.
- The author-ids health check ignores an error older than the last successful pass, and the
connection drops such an error.
- Contributors (page and CLI) shows the counts, override and tier of the account the login is
bound to; an allow that does not count for it shows as not in effect.
The PR author never ranks as a maintainer. Only a maintainer's comment or review
newer than the PR's previous review, or an unresolved, not outdated thread a
maintainer started, must fit; they have a budget of their own, and only such an
entry that still does not fit makes a take unsure.
A duplicate member or kept PR no comparison read a head of is released, or ends
the decision, without comparing again. The base branch a review's diff read
returns is written to the mirror at once. Reviews and pre-merge reviews stored
before bases were recorded take the PR's base once at the upgrade.
- the diff fetch reads the run steps and local actions of .github/workflows at the PR's base, cached per base commit, and stores them beside the diff
- a changed file named in a run step, or inside a local action's directory, is CI configuration; with the workflows unread, script-like paths are
- sourcePaths and parseTriage take the build scripts the base commit's manifests name
- The CI-fix guard applies the patch with git and judges git's own diff and the files before and after it, so a hunk header that names other lines cannot hide a change to test code. Attributes that switch off a lint or a test, and cfg! calls, are read from the patched file's Rust tokens, however they are split over lines or commented.
- Conflict resolutions: a Rust answer is read as tokens; a line on neither side may add no item word, attribute, unsafe code or item macro, and the answer may hold no more of them than its two sides in their own files, so uncommenting a side's code counts as new. Manifest answers keep each line in the table and under the keys it had in its side, and each side's lines in order. A file whose head or base has marker-shaped lines is refused.
- One trust-path rule (build scripts of either parent, CI and tool configuration, agent-instruction files) is shared by every model-patch lane, conflict resolution included.
- Every lane reply parser reads through the strict single-value reader; a fenced reply with a second answer (a value with the same keys) outside the fence is void.
- When the first reply is used after its fix turn is skipped, fails or cannot be merged, it gets every lint check, the model checks the first lint put off included. A fenced first reply can take the fix turn's fields.
- Cargo.toml: the head text of an edited manifest is fetched with the diff and kept beside it; the manifest is parsed whole before and after, and a change to the build script, links, proc-macro, build dependencies, patch or replace, a git or registry source, or a path leaving the repository needs a person
- lockfiles: a source outside the default registry, a new checksum for a locked version, a lockfile change without a manifest change, or a renamed, copied or deleted lockfile needs a person; registry version updates do not
- the trust paths the fix lanes refuse (.github, toolchain, task runner, .gitmodules, CODEOWNERS, agent files) and submodules, symlinks, mode changes and non-image binaries need a person
- diff sections carry both names of a renamed file and decode quoted names; a name that cannot be read needs a person
- docs links are resolved as a browser does: lines starting with ++, schemes without slashes, split attribute values, dot segments
- reviews are shown test files while the diff fits; tests cut for size need a person
- the rule's size is the line count of the cached diff at the reviewed head
- a pre-merge review row without an unsure value reads as unsure
- build scripts are read with a TOML parser in every lane, and the security-fix lane resolves them from the scanned commit