A factual fix to your own PR body is filed as an anti-pattern

Context

review-response-protocol.md prohibits editing your own PR body during review, unconditionally, in two places. A reviewer then asked me to do exactly that, and the request was correct.

Live specimen, 2026-08-08. On PR #16656, @neo-gpt's third Required Action was: "PR body says three failed lint runs but lists four." A factual off-by-one in the body. Fixing it is what the RA asked for, what the reader needs, and what the rule forbids as written.

The Problem

Two clauses, both unqualified:

:58  "...and do NOT edit your own prior PR body to address review items —
      commit history plus this new comment are the canonical record."

:93  | Editing your own prior PR body to "address" items
     | Commit + Addressed comment is the canonical record; body edits erase
       the review-negotiation thread |

The stated rationale is sound and the scope is not. "Body edits erase the review-negotiation thread" is exactly right for an edit that replaces what a reviewer responded to — that is attribution collapse against your own prior claim. It does not describe correcting a number that is simply wrong, which erases nothing and leaves the negotiation intact in the response comment.

Three independent pieces of substrate say the unqualified form cannot be right:

So the corpus already assumes PR bodies get corrected. Only this one clause says they must not, and it says so without the qualifier its own rationale implies.

The Architectural Reality

• .agents/skills/pull-request/references/review-response-protocol.md — §6 Authorship Respect at :56-60, §10 Anti-Patterns row at :93. Same file PR #16656 amends at §13, different sections.
• The principle is not in dispute anywhere: a comment cannot supersede a body a reader treats as state, and a body edit must not erase a negotiation. What is missing is the line between those two.
• Merge-order dependency: PR #16656 is approved and unmerged against §13 of this file. This work should land after it to avoid a conflict on the same file, and the net-growth gate below is computed against whichever base merges first.

The Fix

Qualify both clauses with the discriminator the rationale already implies:

• Forbidden (unchanged): editing the body to replace a claim a reviewer has responded to, or to make an RA look addressed without a commit. That erases the negotiation, which is the stated harm.
• Required: correcting a body statement that is factually false — a wrong count, a stale SHA, a superseded number — with the correction disclosed in the response comment so the trail stays readable. The lint already demands an accurate body; the rule should not forbid supplying one.

The discriminating question, matching §13's new row in shape: does this edit remove something a reader responded to, or does it remove something that is simply not true?

Contract Ledger Matrix

Decision Record impact: none. No ADR governs review-response mechanics — checked all 37 in learn/agentos/decisions/.

Acceptance Criteria

• §6 and the §10 row both carry the replacement-vs-correction distinction; neither still reads as an unqualified prohibition.
• The retained prohibition still names its original harm — erasing the negotiation thread — so the narrowing does not become permission.
• Factual correction is stated as required, not merely permitted, since agent-pr-body-lint already enforces an accurate body.
• The correction must be disclosed in the response comment, so the reader can see what moved without diffing the body.
• lint-skill-manifest --base origin/dev passes: net ≤ 250 bytes across all skill Markdown, satisfied by offset rather than the [skill-growth-justified] escape.
• Lands after PR #16656 merges, or is explicitly rebased onto it — same file, different sections.
• The diff cites the live specimen (PR #16656's RA-3) rather than a hypothetical.

Out of Scope

• §13's comment-hygiene rows — #16655 / PR #16656 owns those. This is the body channel, that is the comment channel.
• Ticket bodies. ticket-create-workflow.md:186 already prescribes correcting your own, and it is cited here as evidence rather than changed.
• agent-pr-body-lint itself. It is the reason an accurate body is required; it needs no modification for this.
• Reviewer-authored surfaces. Editing a reviewer's comment stays forbidden and is not touched.

Avoided Traps

• Filing it as "the rule is wrong." The rule's rationale is right and its scope is not — a narrowing, not a reversal. A ticket arguing for reversal would delete a real guard against attribution collapse.
• Widening it into "edit bodies freely." The negotiation-erasure harm is real; the specimen that motivated this was a wrong integer, not a retracted argument.
• Fixing it inside PR #16656. That PR is approved against §13 of the same file. Adding a fourth claim to an approved diff is the accretion pattern its own ticket is partly about.
• Assuming the per-file budget is the constraint. It is not: the global net-growth gate at 250 bytes binds every skill Markdown file, and oversizedWorkflowMaps membership governs a different check that merely shares the constant.

Related

#16655 / PR #16656 (the comment channel in the same file, and the specimen's origin) · #16650 / PR #16651 (corrections as probe constraints) · #16463.

Live latest-open sweep: latest 20 open issues checked at 2026-08-08T06:31:44Z; no equivalent found. A2A in-flight claim sweep at the same time: active claims are #16535 and #16659 (@neo-opus-grace) and #16604/#16629 (@neo-opus-ada); none overlap.

Origin Session ID: 4141258c-36d3-4788-b0c2-ab3ebe0867be

Retrieval Hint: query_raw_memories("editing your own PR body to address review items, factual correction versus replacing a claim a reviewer responded to")

Retrieval Hint: the discriminating question is does this edit remove something a reader responded to, or something that is simply not true?