fix(iteration): bail on CLOSED PR + fetch before ahead-count#120
Merged
Conversation
…-caught) Bug discovered dogfooding the loop on Titan #1104 (audit retention): - Iteration loop attempt 2 → dirty_no_commit (commit brief) - Iteration loop attempt 3 → committed_not_pushed (push brief) - worker_iterations_exhausted with final_state=committed_not_pushed - PR was real and OPEN at one point, then CLOSED externally; the iteration loop never noticed and kept trying to push to a branch whose work had been abandoned. Two root causes: 1. ``probe_worker_state`` checked ``state == "MERGED"`` but fell through to the "no PR" path for ``state == "CLOSED"``. That fed the COMMITTED_NOT_PUSHED → push brief loop forever. 2. The ahead-count (``git rev-list origin/branch..HEAD``) was measured against a stale tracking ref. A successful push from a prior session still read as 'commits ahead' because nothing in the probe path refreshed ``origin/<branch>``. Fixes: - New WorkerState.CLOSED_PR_ABANDONED. Added to TERMINAL_STATES so the iteration loop bails on it. - Probe returns CLOSED_PR_ABANDONED when the PR is in CLOSED state (not MERGED, not OPEN). - run_iteration_loop main body calls escalate_to_human (label loop:needs-human + diagnostic comment) when it sees the abandoned state, so the dispatcher doesn't keep picking up the issue every tick. - ``git fetch --quiet origin <branch>`` runs before the ahead-count. Subprocess failure is suppressed — degrades to legacy behaviour rather than crashing the probe. Tests: tests/test_iteration_closed_pr.py — 8 new: - CLOSED PR returns CLOSED_PR_ABANDONED (the headline regression pin) - CLOSED_PR_ABANDONED is terminal - brief_kind_for(CLOSED_PR_ABANDONED) is None (no follow-up brief) - MERGED still takes precedence over the new CLOSED branch - OPEN classification unchanged - Probe calls git-fetch BEFORE git-rev-list (ordering pin) - Fetch failure non-fatal (network down doesn't crash probe) - run_iteration_loop short-circuits + escalates on CLOSED_PR_ABANDONED (no dispatch_worker call, escalate_to_human fires) Also updated tests/test_iteration_probe.py::test_terminal_set_... to include CLOSED_PR_ABANDONED in its membership assertions. Full suite: 766 passed (was 758), 0 regressions vs trunk baseline.
This was referenced May 28, 2026
hadamrd
added a commit
that referenced
this pull request
May 28, 2026
… (#139) Dogfood the manifestos system on forge-loop itself by writing the seed quality and testing manifestos that every future forge-loop change is gated against. quality-manifesto.md codifies five rules drawn from this week's persistent-worker work: no shared module-level state (#100), typed Protocol+Fake at every I/O boundary (#104), single Settings source of truth (#98), typed events instead of untyped **fields (#99), and no subprocess.run for SDK-able services (#103, #105). Each rule names the concrete issue it came from so future contributors know the *why*. testing-manifesto.md codifies six rules drawn from this week's iteration-probe bugs: one test per state-machine edge plus a fallthrough adversarial (would have caught #97/#120/#128), an adversarial test for the false case of every external-dep assumption, both ==0 and !=0 branches for every subprocess.returncode (specifically #128), a contract test pinning every Fake to its Real, hypothesis property tests on >4-branch / user-input functions (#102), and an adversarial test that every infinite-loop guard actually fires. tests/test_manifestos_discovery.py is the meta-validation gate: it discovers and parses both files, asserts each rule has a rationale, asserts the spec-mandated issue references are present, and includes adversarial tests that stubs and missing files are detectable. 22 tests, all pass.
5 tasks
hadamrd
added a commit
that referenced
this pull request
May 28, 2026
…loop) (#149) Closes the feedback loop the CTO described: every bug we fix becomes a permanent gate. Today's PR #147 (critic SDK event-capture mismatch) exposed a 4-PR train of bugs with the same shape — #97, #120, #128, #147 — all driven by string-literal discriminators that didn't match across module boundaries. The critic (PR #141) reads the quality manifesto + flags sev1 violations. This rule + the critic infrastructure together mean the next worker that writes ``event["type"] == "result"`` (or similar cross-module string-comparison) gets the PR auto-blocked with the manifesto rationale.
hadamrd
added a commit
that referenced
this pull request
May 28, 2026
Adds the customer-facing documentation for the manifestos + brainstormer feature that closed the cosmetic-tickets gap. Real customers consuming this OSS need to know: 1. The four files they own (.forge/product-vision.md, axes.yaml, quality-manifesto.md, testing-manifesto.md). 2. The brainstormer dry-run + --apply workflow. 3. The feedback loop (`forge-loop manifesto suggest --from-pr <N>`) where every bug becomes a permanent gate. 4. What the worker + critic see (manifestos injected into briefs; sev1 violations block auto-merge). README: new section "Manifestos & the brainstormer (axis-aligned tickets)" between Briefs and CLI reference. CLI reference table gains `brainstorm`, `brainstorm --apply`, `manifesto suggest --from-pr`. GUIDE: new section 4 "Manifestos: drive what gets built (not just how)" between "discipline matters" and "the brief is your contract" — with the real Titan brainstormer output as the worked example. Sections 5-10 renumbered accordingly. Both docs cite PR #147 as the canonical feedback-loop example: a stringly-typed event-boundary bug that surfaced after #97/#120/#128 all had the same shape; the fix landed the manifesto rule that the critic now enforces.
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
1 participant
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.
Bug
Caught dogfooding the loop on Titan #1104:
Root cause
Fix
Test plan