Idempotent improved-wake publish (last activation-blocker)

[renmengye]

The last dispatcher activation-blocker. A wake that opened the PR but died before recording it left the run WAITING; the re-wake re-pushed non-fast-forward and opened a duplicate (or recorded ABORTED over a live PR).

resume_run now asks find_open_pull_for_head (new GitHubClient method — GET /pulls?head=owner:branch&state=open) before publishing, and if a PR is already open for the run's head, reconciles the record to in-review on that PR — no re-push, no duplicate. Best-effort: a lookup failure falls through to the normal publish.

Test: a re-wake with an existing open PR reconciles (no new PR, not re-armed, in-review, snapshot released). Full gate green, 672 tests.

With this in, the dispatched path can be turned on (AUTORESEARCH_DISPATCH_WAKE).

🤖 Generated with Claude Code

[github-actions]

Round 1 — reviewed head c83bbab1 — reviewer hermes/gpt-5.6-terra.

second opinion — terra
Advisory findings from autoresearch — the code owner decides. Reply to disagree; the autoresearch:no-review label opts this PR out.

Verdict: nothing blocking — 1 advisory note.

1 finding attached to the lines below.

[github-actions]

Round 2 (re-run on the same head) — reviewed head c83bbab1 — reviewer hermes/gpt-5.6-terra.

second opinion — terra
Advisory findings from autoresearch — the code owner decides. Reply to disagree; the autoresearch:no-review label opts this PR out.

Verdict: nothing blocking — 1 advisory note.

1 finding attached to the lines below.

[github-actions]

Round 1 — reviewed head c83bbab1 — reviewer claude/claude-opus-5.

Advisory findings from autoresearch — the code owner decides. Reply to disagree; the autoresearch:no-review label opts this PR out.

Verdict: 1 blocking, 3 advisory.

3 findings attached to the lines below.

Advisory (non-blocking):

• New test does not check the workspace branch after reconcile (tests/test_climb.py:2404; medium)

The broad except Exception around find_open_pull_for_head also swallows AttributeError, so any test double lacking the new method silently falls through to the normal publish; that is intended best-effort behavior but hides interface drift. I could not run the test suite (read-only checkout), so the claim of 672 passing tests is unverified. Whether a follow-up push to the default branch would actually land depends on the target repo's branch protection, which is outside this context.

[github-actions]

Round 2 — reviewed head c6b4bb26 — reviewer claude/claude-opus-5.

Advisory findings from autoresearch — the code owner decides. Reply to disagree; the autoresearch:no-review label opts this PR out.

Verdict: nothing blocking — 5 advisory notes.

1 finding attached to the lines below.

Advisory (non-blocking):

• Reconcile skips the issue comment that reports the finished run (src/autoresearch/climb.py:610; medium)
• A failed lookup still leads to the duplicate/abort it is meant to prevent (src/autoresearch/climb.py:591; medium)
• Reconcile writes no run report (src/autoresearch/climb.py:605; low)
• Question: Reconcile does not put the workspace on the PR branch (src/autoresearch/climb.py:597; low)

I could not run the test suite (read-only checkout, no execute tool), so the "672 tests green" claim is unverified. The new test does exercise the reconcile branch genuinely: with the check removed, FakeGitHub.create_pull would append and github.prs == [] would fail.

[renmengye]

False positive on the current head (c6b4bb2): the reconcile DOES arm auto-merge — see climb.py where it computes pr_number from the adopted PR and calls arm_auto_merge_when_review_required when the PR is not a draft, right after the checkout. test_resume_improved_reconciles_to_an_existing_pr asserts github.armed == [("org/pilot", 7)], i.e. the adopted PR #7 is armed. Resolving as addressed.