You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Re-read each issue and pull request rather than trusting these summaries.
repo-worktree's hook step installs into the git hooks dir every worktree shares #1816, the worktree hook-install defect.repo-worktree's hook step installs into the git hooks directory that every linked worktree shares, which broke Financial-Modeling's commits. Done looks like the skill checking the common hooks directory first and never installing from a linked worktree, plus gh-write-guard denying a commit-time core.hooksPath override.
unattended-handoff: adopted lesson issues stall, and the promotion STOP matches any decision issue #1812, fixed on reasoning rather than waiting for a live run to hit it. The live run exercised neither path, and a natural run is unlikely to soon. For an adopted lesson, count an answer recorded on the issue as closing its choices. For the promotion STOP, match only a decision issue that states it blocks the open develop -> main PR.
Exercise the unattended loop's parked-lane half. An attended "resume the handoff" of a lane parked under blocked has never run. Done looks like one parked lane coming back through the attended procedure intact.
Say the Hub Commits the Report Wherever Done-Means-Audited Is Stated #1826: reword the remaining places that say the audited repository "carries" a committed reports/<repo>/audit.md. Those are GOVERNANCE.md "Repository Onboarding and Conformance", AUDIT.md section 1, STANDUP.md section 5, standup-a-repo, docs/reusable-workflows.md, and TODO.md. The hub is the one that commits the report.
Downstream resyncs for the new Fleet Bootstrap text. 2.0.680 changed the byte-locked "Fleet Bootstrap" section, so every carrier reports drift on it until its next resync. When aiopurpleair resyncs, its handoff (aiopurpleair Update codegen files #177) should drop step 3, "commit an audit report", because the report is the hub's to commit.
Financial-Modeling's shared pre-commit hook is broken until the maintainer restores .git/hooks/pre-commit.legacy over the prek hook, a one-line mv in the primary checkout that the auto-mode classifier correctly refuses an agent. Until then, every commit there fails. repo-worktree's hook step installs into the git hooks dir every worktree shares #1816 is the hub-side cause. Carried.
Downstream promotions: the version-literal removals sit on develop in eleven fleet repositories, and each one's main carries the literals until the maintainer promotes it. Carried.
A native Windows host is needed for step 8's first item. Carried.
CodeRabbit needs an explicit @coderabbitai review here and allows about one review an hour. Carried.
Empty. No open issue carries the decision label. This round put these questions to the maintainer and recorded each answer:
Keep AUDIT.md section 8 as it stands. A downstream repository never opens a hub pull request to write its own audit report, files hub findings as issues, and may open other hub pull requests against develop. Fix the wording that contradicted it.
A downstream repository's own session resyncs itself per RESYNC.md, and resync-a-repo from a hub checkout is the other route.
Resolve the resulting check-this-repo contradiction in the same pull request.
Settled audit-report authorship and downstream self-resync (State That the Hub Commits Audit Reports and a Downstream Session Resyncs Itself #1823). The carried Fleet Bootstrap told every repository to "commit the report" without naming the hub, so downstream agents kept asking the maintainer for hub-write permission. The Fleet Bootstrap, the host fleet block, RESYNC.md, AUDIT.md, check-this-repo, and resync-a-repo now agree.
Two sessions each driving toward a promotion. This round one promotion PR was agreed by cross-session message before either session opened one. A run at scope main or release alongside an attended session driving a promotion needs the same agreement before it starts.
An unattended run feeds on its own findings. A worker files a pre-existing gap and the next picker takes it, which can drift the run away from the maintainer's priority order. Check the picks against this list.
Chasing a prose-gate failure on a line this branch never touched. CI checks the merge commit against the branch's merge base, so a comment line develop gained after the fork shows up as this branch's. Merge develop into the branch rather than editing the line.
Stating one rule in a hub-only document while carried text says the opposite. A rule a downstream agent must follow has to be in what that agent reads. AUDIT.md section 8 was right the whole time, and the byte-locked Fleet Bootstrap was what agents actually read.
New learnings
Copilot can write findings=N in its coverage marker while posting no inline comment and no suppressed block, with the only stated finding in the overview headline. Read the headline as the finding and dispose of it in a PR comment. This round's headline, on Promote the Handoff Loop, Version-Literal Ban, and Audit Report Authorship to Main #1827, was declined with evidence.
The attended resume of a parked lane has not been exercised yet (step 3).
Next steps, in priority order
Re-read each issue and pull request rather than trusting these summaries.
repo-worktree's hook step installs into the git hooks directory that every linked worktree shares, which broke Financial-Modeling's commits. Done looks like the skill checking the common hooks directory first and never installing from a linked worktree, plus gh-write-guard denying a commit-timecore.hooksPathoverride.blockedhas never run. Done looks like one parked lane coming back through the attended procedure intact.reports/<repo>/audit.md. Those areGOVERNANCE.md"Repository Onboarding and Conformance",AUDIT.mdsection 1,STANDUP.mdsection 5,standup-a-repo,docs/reusable-workflows.md, andTODO.md. The hub is the one that commits the report..github/skillstree into the seven repositories listed there, so Copilot states coverage on their pull requests.bootstrap.ps1, then close Bootstrap registers its temporary tree as the live Claude Code skills source, then deletes it #1756; Keep an Owned Tree Recognizable When Its Final Directory Removal Fails #1795, then Name the Leftover Old Tree When the Bash Loader Cannot Remove It #1790, Drive the PowerShell Loader's Kept-Tree Functions from the Test Suite #1791, Lock the Bootstrap Directory for a Kept-Tree Run #1792; Gate Every Repository's Python on Its Code, Not on Its Config #1800 (settle its three decisions with the maintainer before any edit, with Spec: python.profile.detect does not describe a virtual uv workspace root #1778 folded in); The menu's skills task registers a clone it then removes #1771, then Re-point a Claude Code registration an older bootstrap left dangling #1769; Say Where the Deploy-Site Task Actually Requires the SSH Key #1801; Declare the Default Workflow Token Permission in the Repository Settings Payload #1783; A regex pins a phrase inside a # comment, so rewrapping that comment fails the test with a false explanation #1732, then skills_install --report follow-ups: stamp-less live report, bootstrap remedy text, --intended guard test #1759, skills_install source_ref hides ignored files and trusts an enclosing git repo #1758, skills_install --report's live channel does not check that the plugin is installed and enabled #1757; A stacked pull request's diff reopens after the parent squash-merges, and nothing documents it #1742, Session Handoff [default]: One Issue Closed Whose Premise Was Wrong, and a Release #1735, Converge Financial-Modeling with the hub baseline (32 findings) #1579, The widened configuration shape misses the unspaced key=value an environment file writes #1717, issue-ref reads no URL outside instruction text, so an own-repository tracker URL in code is banned and unreported #1719; lower, Two skills describe the live skills channel only as a git checkout #1772, A kept skills tree edited by hand still stamps clean #1775, Nothing tests the bootstrap loaders' kept-tree swap sequence #1767, Two concurrent bootstrap runs can remove each other's tree #1764, install-skills.ps1 fails under a directory whose path contains a wildcard character #1765.External blockers
.git/hooks/pre-commit.legacyover the prek hook, a one-linemvin the primary checkout that the auto-mode classifier correctly refuses an agent. Until then, every commit there fails. repo-worktree's hook step installs into the git hooks dir every worktree shares #1816 is the hub-side cause. Carried.@coderabbitai reviewhere and allows about one review an hour. Carried.Internal dependencies
State
The parked decision queue
Empty. No open issue carries the
decisionlabel. This round put these questions to the maintainer and recorded each answer:AUDIT.mdsection 8 as it stands. A downstream repository never opens a hub pull request to write its own audit report, files hub findings as issues, and may open other hub pull requests against develop. Fix the wording that contradicted it.RESYNC.md, andresync-a-repofrom a hub checkout is the other route.check-this-repocontradiction in the same pull request.What the last round did
unattended-handoffat scope develop with a two-round cap. Both rounds ran picker then worker and returned well-formed lines: DONE Session Handoff [auto-1725]: Bound the Fleet-Review Coverage Marker's Indentation (#1725) #1818 via Refuse a Coverage Marker Indented Into a Code Block #1820, then DONE Session Handoff [auto-1821]: Strip a Tilde or Unclosed Fence Before Reading Review Coverage (#1821) #1824 via Strip a Tilde or Unclosed Fence Before Reading Review Coverage #1825. The workers took about 43 and 28 minutes. Nothing parked, and no lesson or decision issue was filed. Round 1's worker filed Strip a Tilde or Unclosed Fence Before Reading Review Coverage #1821 as a pre-existing gap, and round 2's picker took it.RESYNC.md,AUDIT.md,check-this-repo, andresync-a-reponow agree.What not to repeat
AUDIT.mdsection 8 was right the whole time, and the byte-locked Fleet Bootstrap was what agents actually read.New learnings
findings=Nin its coverage marker while posting no inline comment and no suppressed block, with the only stated finding in the overview headline. Read the headline as the finding and dispose of it in a PR comment. This round's headline, on Promote the Handoff Loop, Version-Literal Ban, and Audit Report Authorship to Main #1827, was declined with evidence.