Repository navigation
π changelog: recover 14 fragments stranded in src/changelog.d and fail the release on strays - #18360
Conversation
β¦l the release on strays Fourteen fragments from PRs merged 2026-10-20..27 were written under src/changelog.d/, which compile-changelog.sh never reads, so their entries never reached CHANGELOG.md. Move them to the repo-root changelog.d/ (one needed the required '- ' bullet prefix) so the next release carries them, following the #11928 / #12675 precedent. Add a release-time backstop: compile-changelog.sh now fails loudly while any fragment other than README.md sits in src/changelog.d/ next to the changelog, so a stray that slips past the PR guard turns the next release job red instead of vanishing. Tests cover the README-only, stray-only and mixed cases plus a repo invariant that src/changelog.d/ holds only its README. Spell out the ../changelog.d/ path for agents working inside src/ in AGENTS.md, both READMEs and the scoped knowledge entry. Fixes #18264 Signed-off-by: Andy Anderson <andy@clubanderson.com>
|
Hive status β updated in place; edits do not notify. Full history is in the Activity Monitor. @clubanderson this PR is held at Why:
Next: the hive does not push to your branch. Push a new head (or reply here); CI and the review swarm re-run automatically and the PR returns to the normal path. Review findings from Routing, not a defect. The change looks correct to me, but it edits the pinned verifier tree in What I verified:
Open point (info): the pin is a literal over the whole |
Signed-off-by: Andy Anderson <andy@clubanderson.com>
There was a problem hiding this comment.
Decision needed: @clubanderson should confirm the re-pinned TRUSTED_SCRIPTS_TREE in .github/scripts/integration-governance.cjs is acceptable, since it is a trust-surface edit that a person must sign off.
I read the change and found no defects. It touches a trust surface, so I am routing it rather than approving.
Trust surface. .github/scripts/integration-governance.cjs:12 changes the pinned TRUSTED_SCRIPTS_TREE. The constant is compared against rev-parse <source>:src/scripts at lines 98 and 102. This PR edits src/scripts/compile-changelog.sh and test-compile-changelog.sh, so the pin has to move. I checked the new value against the head commit (20df9a2): the src/scripts tree there is b60380f8d0040628eba73ad17c108b1be87c7fdd, which matches the new pin. It is the only .github file in the PR. The PR body does not mention the re-pin, though commit c26ad86 does.
Code.
- The backstop in
compile-changelog.shruns before the no-op exits and skipsREADME.md. It resolvessrc/changelog.dnext to the changelog path, so the fixture runs in a temp dir don't see the real repo. - Case 5d covers three situations: README only, a stray with no root fragments, and a stray beside valid root fragments. In the last two it checks that nothing is consumed.
- The repo-invariant check looks at
$script_dir/../changelog.d. With the script insrc/scripts, that resolves tosrc/changelog.d, as intended. - The 14 renames are 100% similarity. The one file that needed a
-prefix has it.
I did not run anything locally (per AGENTS.md); CI is green on the listed checks.
β hive: agent=reviewer-2 backend=claude model=claude-sonnet-5-5 claude=2.1.325
|
[APPROVALNOTIFIER] This PR is APPROVED Approval requirements bypassed by manually added approval. This pull-request has been approved by: The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
Fixes #18264
Why
src/scripts/compile-changelog.shcompiles only the repo-rootchangelog.d/. Fourteen fragments from PRs merged between 2026-10-20 and 2026-10-27 were written tosrc/changelog.d/instead.changelog-fragment-guardfailed on those PRs, but the skipped label-event runs that followed counted as passing (fixed by #14767 on 10-28), so they merged. None of the 14 entries, and none of their PR numbers, appear anywhere in v5'sCHANGELOG.md. Thanks to the reporter for the careful inventory.What
git mvall 14 tochangelog.d/so the next release compiles them. This follows the precedent of π± chore(changelog): move orphaned fragments from src/changelog.d/ to changelog.d/Β #11928 and π changelog: move misplaced fragments from src/changelog.d to changelog.dΒ #12675, which moved strays forward rather than backfilling released sections.fixed-supersession-sweep-stage-machine.mdstarted with prose, so it gets the required-prefix. The content of the other files is unchanged. PRs: π Adoption: a hive-adopted PR that conflicts with its base gets rebased by hiveΒ #13280, π cost: scan per-agent Copilot session trees and split cached inputΒ #13486, β¨ acmm: reviewer agent joins the L3 and L4 packsΒ #13609, π claims: contributor disconnects release the issue and flapping no longer starves hive agentsΒ #13758, β¨ dashboard: activity monitor names the sentinel alert in the merge-gate laneΒ #13760, β¨ dashboard: Accept triage in the activity monitor row menuΒ #13805, π Route supersession sweep escalation through the stage engine and let the hive resolve its own superseded PRsΒ #13807, π watchdog: widen the no-production window to the agent's own cadenceΒ #13821, π ci gate: ignore held runs whose workflow already passed on the head and name same-repo bot PRsΒ #13822, π± crux: Runtime seam for the sidecar plan, stage 1Β #14208, π± cmd/hive: consume the agent runtime through the crux.Runtime seamΒ #14220, π Hub holds upgrades the spoke is draining instead of orphaning themΒ #14322, β¨ Live per-run cost Β· tokens Β· elapsed Β· tests Β· files line on agent cardsΒ #14580, β¨ Credit issue authors on clanker-opened PRs at merge timeΒ #14585.compile-changelog.shnow fails loudly, before its no-op exits, while any fragment other thanREADME.mdsits insrc/changelog.d/next to the changelog. That directory is resolved relative to the changelog path, andSTRANDED_FRAGDIRcan override it. The PR guard already rejected such fragments (π fix(changelog): reject fragments under src/changelog.d/ and point authors to the repo rootΒ #11926). The gap was that once one slipped past the guard, nothing noticed it again. Now the next release job turns red, which is the same "fail on the very next merge" approach as therelease: nonerejection (π± chore: reuse off-prompt pane fixture in testΒ #12928).test-compile-changelog.sh(case 5d):README.mdalone is fine;CHANGELOG.mdbyte-identical;src/changelog.d/holds only its README.AGENTS.md(whichCLAUDE.mdlinks to), both READMEs and the scoped knowledge entry now spell out that from insidesrc/the path is../changelog.d/. During the window when these were written,AGENTS.mdsaid onlychangelog.d/<...>.mdand did not say "repository root". The root wording arrived with π± AGENTS.md: keep tone and workflow rules; move repository facts into scoped knowledge entriesΒ #17051 on 11-07.v6
v6 has 81 strays. 13 of them are these same v5 files, and the v6 top-up carries their move. The 68 that exist only on v6 are handled in a separate v6 PR. That PR must merge before the top-up carries this backstop, or v6's next release will go red on the leftovers. That is the intended behavior.
π€ Generated with Claude Code