pr-review-report: sanctioned PR-body edit subcommand — producer deadlocks on body-only vetter rejects (QA §8 block)

[thedavidmeister]

🤖 ai:producer

The vetter now rejects PRs solely for a missing QA-GUIDE §8 evidence block in the PR BODY (and for stale body claims), but the producer has no sanctioned path to mutate a PR body: gh pr edit is forbidden and pr-review-report has no body subcommand. This deadlocks PRs in ai:reject with no producer-side exit:

• docs(setup): fix README setup steps and add missing env var cyclofinance/cyclo.site#385 — vetter (twice): 'coverage itself now looks complete incl. the root npm ci from the rework note — add the [QA §8] block and re-push'; the code rework is done, only the body block is missing.
• test: assert authoringMetaV2 word names and descriptions rain.flare#130 — newest vetter reject includes 'missing QA-GUIDE §8 evidence block' alongside code grounds (code grounds are being reworked; the body ground remains unservable).
• fix: guard weighted-average merge against zero outputAmountSum in historicalOrderCharts raindex#2777 — rework fully applied and mutation-evidenced; the PR body's stale 'unweighted mean' claims cannot be corrected (producer staged a corrected body at /home/gildlab/code/pr2777-body-corrected.md for a human paste).

Proposal: a guarded subcommand, e.g. pr-review-report set-qa-block <owner/repo> <pr> --body-file <file> that (a) only operates on OWN open PRs, (b) only appends or replaces the delimited QA §8 section (never the Closes lines or prose above it), and (c) records the edit in a trusted marker comment — keeping body mutation constrained while unblocking the reject→rework→re-vet loop. A sibling flag (or the same command) could patch demonstrably-false body claims flagged by a rework note.

Filed by the producer routine; these PRs are being flagged ai:blocked-infra referencing this issue until a sanctioned path exists.

[thedavidmeister]

Consolidating #112 into this issue — it was filed 2026-07-28 without spotting this one, and it is the same ask. Three things from it are worth carrying over rather than losing.

The deadlock has grown, with a number

pr-review-report --help still has no body-writing subcommand, and campaign-settings.json still denies Bash(gh pr edit:*). As of 2026-07-29, 56 of 100 open PRs carry no ## QA block. Every one is a candidate for a reject only a body edit can clear. Named: rain.factory.deploy#4/#5/#6, rain.solmem#100/#101, rainix#289, rain.deploy#19, cyclo.sol#56.

Worth being precise about the cost: in this class the code is not the problem. The vetter's own notes certify the diffs sound — "verify clean locally", "test code audits sound: all 11 named tests present and verified discriminating", "Code itself audits sound". PRs are parked over a missing paragraph.

Two properties to design in rather than discover

It should APPEND the evidence block, not rewrite the body. The failure being fixed is a missing section, not a wrong body. A subcommand that replaces the whole body invites losing whatever the author wrote, and makes a small repair indistinguishable from a rewrite in the PR's edit history.

It should validate what it writes. require-qa-block already knows what a well-formed section 8 block looks like — it is the gh pr create gate. A body-writing subcommand that produced a body that gate would refuse is a bug discoverable only at the next open. Reuse the predicate rather than re-implementing it.

The half that is already solved, and the half that is not

#83 (closed) fixed the create side: require-qa-block refuses a gh pr create whose body lacks the block, so the population of stuck PRs stops growing. It does nothing for a PR already open — the gate is on create, not edit, and there is no retrofit path. That retrofit is what this issue is.

Also worth deciding, separately

Should the vetter's reject for this be a reject at all? A missing evidence block is not a code defect, and review-prompt.txt is explicit that a reject note "MUST name a specific code defect that would fail review EVEN IF CI WERE GREEN". Rejecting for a missing paragraph stretches that, and it collapses "your code is wrong" with "your paperwork is missing" — the human queue never sees code the vetter already certified.

Now that #83 prevents the omission at source, this may be a rule that has outlived its cause. It is a separate decision from the subcommand and should not block it.

Check

A PR rejected solely for a missing QA-GUIDE section 8 block can be repaired by the producer in one sanctioned call, the repaired body passes require-qa-block, and the rest of the body is byte-identical to what it was.