Session Handoff [review-tooling]: Incidental Counts Rule Landed, Copilot Overview Reader Gaps Next #3171

Next steps, in priority order

1. Read Copilot's "Previously Missed Issues" Overview Block in pr_review.py #3167, then Read Copilot's "Open Issue" Overview Block in pr_review.py #3170: teach pr_review.py the Copilot overview shapes it cannot read. Read Copilot's "Previously Missed Issues" Overview Block in pr_review.py #3167 covers the "previously missed issues" block, which holds real findings on code the head did not change. It is higher stakes, because the digest counted nothing and a session trusting it could merge with unanswered findings. Done looks like: wait and status count each previously missed FILE:LINE entry as a finding on the head, like a suppressed finding, instead of exiting 43. Read Copilot's "Open Issue" Overview Block in pr_review.py #3170, the "open issue" block, mostly restates threads already counted, so done is recognizing it as a restatement. The maintainer asked for these, and similar reader gaps, in the next session.
2. Triage the related pr_review.py reader issues in the same pass, and fold each into that change where it touches the same parser: pr_review.py Passes a Clean-Verdict Copilot Round Whose Headline Sentence Names a Finding #2567 (a clean-verdict headline that names a finding) and pr_review.py: a GitHub Actions billing-failure notice can be misread as UNRECOGNIZED_REVIEWER_OUTPUT #1071 (a billing-failure notice misread as unrecognized output). Leave any that need a separate design as their own issues.
3. Finish the Incidental Count Sweep, Including Counted Section Titles #3169: finish the incidental-count sweep, including the counted section titles that need a heading rename. This is lower priority and fine to leave for a later round.

External blockers

None.

Internal dependencies

• Read Copilot's "Open Issue" Overview Block in pr_review.py #3170 depends on Read Copilot's "Previously Missed Issues" Overview Block in pr_review.py #3167's parser change, since both read the same overview body. Land Read Copilot's "Previously Missed Issues" Overview Block in pr_review.py #3167 first, or fix both in one change.

State

• Merged into develop: #3168, squashed as 3bb88965. No develop -> main promotion was opened, by the maintainer's choice of "into develop" for this round. The next promotion carries it, and that promotion PR's body owes Closes #3160, since Refer to an Enumeration Instead of Stating an Incidental Count #3168 used Refs.
• The feature branch, its worktree, and its remote branch were removed. The primary checkout is on develop, clean and fast-forwarded to 3bb88965.
• No release dispatched. No other repository touched.

The parked decision queue

The queue holds 0 open issues with the decision label, counted when this was written. Nothing is parked.

What the last round did

• Added GOVERNANCE.md "Documentation Style Conventions" > "Incidental Counts", carried whole into comment-and-doc-style as a generated include, with a pointer line in fleet-code-review. The rule applies in every fleet repository, as the maintainer asked.
• Removed the incidental counts in AGENTS.md, GOVERNANCE.md, CODESTYLE.md, WORKFLOW.md, the skill sources, scripts/README.md, and the scripts/handoff.py docstring.
• Filed Finish the Incidental Count Sweep, Including Counted Section Titles #3169 for the rest of the sweep.

What not to repeat

• Removing a count orphans its antecedents. The first local strict pass raised 14 findings, nearly all of them a pronoun or a "these" now bound to the wrong noun. The rest were a later "both", "either", "the third", or "all three" in the same paragraph that still assumed the count. Read each edited paragraph whole after the removal, not just the edited sentence. Finish the Incidental Count Sweep, Including Counted Section Titles #3169's body says the same, for the sweep that remains.
• The sweep's suggested rewrites are drafts, not edits. A first-pass agent proposed things like "fails in ways the chain closes" and "Things make that loop hard", which read badly. Judge each one, and keep a count an ordinal structure relies on.
• Copilot's "Needs a closer look" headline can name findings without locating them ("a grammar error"). pr_review.py reported it as VERDICT_NAMES_UNCOUNTED_FINDINGS. The answer was to sweep every rewritten sentence, fix the verbless fragments it found, and reply on the PR with comment. Re-requesting the round would not have located the finding.

New learnings

• The pronoun-fallout lesson is recorded in Finish the Incidental Count Sweep, Including Counted Section Titles #3169's "Possible Direction", where the remaining sweep will read it. No governance text was added for it.