Repository navigation
Merge pstack-claude 0.9.79 into the Herdr fork (0.9.80) - #22
Merged
Merged
Conversation
…ael-denyer#232) * fix(reflect): keep the parent's verdict out of reviewer prompts A focus note beside the transcript path stated the parent's diagnosis as fact, and two of three reviewers built on it (michael-denyer#231). reflect now adds nothing beside a transcript path, and the fallback digest records events in turn order without a diagnosis, verdict, or cause. Release 0.9.76. * Keep reflect's fallback digest to one line per event
The panel tier in models.json gains haiku, so arena, architect, and interrogate run four models when no override sheet names the role. Release 0.9.77.
* feat(pstack): sync to upstream 1e56b29 poteto-help offers /setup-pstack when the model sheet is missing and the answer depends on it. * feat(pstack): sync to upstream df58112 (v0.15.15) Upstream dropped one panel model and lowered its Opus effort. The port keeps its own panel and effort from models.json, and new substitution rules stop an upstream model change from conflicting with forked lines. Release 0.9.78.
) * test(sync): pin the period after a setup-pstack slug * fix(sync): keep the period after a setup-pstack slug The slug rule's last segment admits dots, so a bare slug that ended a sentence took the period into <slug>. A lookbehind stops the match before a trailing dot. * docs(pstack): name the default effort in poteto-help and correct the 0.9.78 sync counts poteto-help offered a reasoning budget, a step the port's setup-pstack does not have. The 0.9.78 entry listed counts that summed to 87 over 82 synced files, so it now states what the two sync commits show.
…ichael-denyer#236) A timer delay above 2^31 - 1 ms is replaced with 1 ms, so --interval above 2147483.647 seconds made every poll sleep return at once. The parser now refuses it. The Lean model of the polling arithmetic comes with the change; its sleep bound assumes a sleep lasts as long as asked, which this keeps true for the CLI.
…enyer#237) /loop treated any non-idle session as a run in flight. Pi is also not idle during a manual compaction, and it emits agent_settled only when a run ends, so a /loop typed during a compaction left the slot sealed and the next run was refused every wakeup. The scheduler now tracks a run from agent_start to agent_settled and seals only then. The TLA+ model of the scheduler comes with the change; two of its runs still fail by design and record narrower findings this does not fix.
…denyer#238) The release read the lock, saw its own pid, and unlinked the lock by path as a separate step. A forced takeover between the two replaced the lock, and the release deleted the replacement, which left the forced writer with no lock on disk. The release now runs its read and unlink behind the claim a takeover holds, and waits for a live claimant to finish. The TLA+ model of the lock comes with the change; two of its runs still fail by design and record the empty lock a filesystem without hard links allows.
Co-authored-by: Alex OG Dev <Alexdev@Overcy-Dev-1.local>
…el-denyer#240) * test(shipping): answer the shipping tests from one fake GitHub Both shipping test files dispatched on the query text themselves, and each compared the dequeue mutation with a fixed string. That check failed a query that differed only in spacing and passed a wrong input field on disablePullRequestAutoMerge. fakeGitHub holds each mutation's required input field in one table and answers the way the endpoint does: a query gets back only the fields it selects, and a mutation needs the right name, input field, variable type, and pull request id. * test(shipping): match only top-level fields and allow a named mutation The field filter matched the nested id in mergeQueueEntry { id }, so the fake still returned the pull request id to a query that no longer selected it. The mutation pattern also refused a named operation, which GitHub accepts.
Takes upstream's haiku in the default panel, reflect reviewer prompts without the parent's verdict, the cursor/plugins pin at df58112 (v0.15.15), the ship-pr dequeue fix, and the watch-pr, Pi scheduler, Pi agent restore, orch store lock, worktree-audit, and sync fixes with their Lean and TLA+ models. Keeps the fork's Herdr dispatcher, --next, Agent gate, routes, Herdr model menu, and the real Herdr CI job. The reflect fork entry takes upstream's description and keeps the Herdr pointer, and the README pin moves to df58112. Upstream released its own 0.9.76 and 0.9.77, so its entries become subsections of this one and the version moves to 0.9.80. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
This merges
michael-denyer/pstack-claude0.9.76 through 0.9.79 (3b0bc62..60ae9e2, 11 commits) into the fork.From upstream
haikujoins the default panel.reflectkeeps the parent's verdict out of its reviewer prompts.cursor/pluginspin moves todf58112(v0.15.15).ship-pr cancel-pendingcan dequeue a PR again.watch-pr(--intervallimit), the Pi scheduler and agent restore, theorchstore lock,worktree-audit, andsync. The Lean and TLA+ models behind them come too.Kept in the fork: the Herdr dispatcher,
--next, the Agent gate, routes, the Herdr model menu inmodels.json, the Herdr pointers in skills, and the real Herdr CI job. Upstream touched none of them.Conflicts resolved
tools/forks.json: thereflectentry takes upstream's longer description and keeps the fork's Herdr pointer.CHANGES.md: upstream released its own 0.9.76 and 0.9.77, which collide with the fork's. Its four entries go under the new## 0.9.80merge entry as### pstack-claudesubsections, as in earlier syncs.VERSIONand the manifests: set to 0.9.80 and regenerated.README.md: the upstream pin moves todf58112. The fork's README test caught the stale pin.Merge with a merge commit, not a squash, so the next sync starts from
60ae9e2.Test plan
bun tools/generate.mjs --checkbun tools/sync.mjs <component> <pinned sha> --dry-runforpstackandcursor-team-kit: both exit 0.node tests/herdr-runtime.test.mjspoteto-mode/scripts: typecheck and Prettier pass. Inbun test orch watch-pr herdr, 331 pass and 1 fails: theorch"Graphite output that carries colour codes" test. It fails the same way onmainon this machine.bun test tests/: 1195 pass and 1 fails: the 5 s timeout inlog.test.mjs"40 concurrent writers". It fails the same way onmainon this machine.🤖 Generated with Claude Code