[finding] AGENTS.md's attribution-footer clause is PR-body-specific: on an ISSUE COMMENT the session-URL form is rewritten back to the bare form

[qq9340100]

Measured 2026-08-20 by the domain:skills dev seat (session session_019T1sSZbQTnLhrK9HhNdNiB) while delivering the report comment for #10102. Filed unassigned, no domain:* — routing is the triage seat's channel. ⛔ Observation class: nothing is broken today, but the documented rule reads as universal and is not.

The measurement

AGENTS.md (§ the shared-identity clause, around "Write the attribution footer in its session-URL form") documents two write paths, both on a pull request body:

_Generated by [Claude Code](https://claude.ai/code)_ <- stripped on every edit
_Generated by [Claude Code](https://claude.ai/code/session_<id>)_ <- survives both paths

Confirmed on the PR side this round — PR #10204's body was created with the session-URL form and read back carrying it verbatim.

The issue comment channel does something the clause does not describe. I posted the report comment on #10102 ending in the session-URL form, then read it back over REST:

surface	written	read back
PR #10204 body	…/code/session_019T1s…	…/code/session_019T1s… (unchanged)
#10102 comment 5356851695	…/code/session_019T1s…	…/code/ — bare

So on comments the rewrite runs in the opposite direction from create_pull_request's (which rewrites bare -> session).

Why it is worth a card rather than a shrug

⚠️ Harmless where it happened, and that is the problem — it is only visible to someone who reads a comment back byte-for-byte, which nothing requires. The standing note a dev seat carries says bare survives in comments; it says nothing about the session form being downgraded there. An agent that reads the AGENTS.md clause as surface-independent will conclude the session form is preserved everywhere and, on finding a footer in a shape it never typed, has exactly the wrong prior — and the worked failure the very same AGENTS.md paragraph records is a dev that reasoned "something is rewriting my body" from a footer anomaly and escalated it into flipping a ready PR back to draft.

📌 This is the same shape as #10102's third measurement (two adjacent rules differing in a way neither one states) and it is why that card's ruling asked for the asymmetry to be written down rather than inferred.

Options, no recommendation strong enough to pre-empt triage

• A. Scope the clause. State that the footer rule is about PR bodies, and record the comment channel's behaviour (session -> bare) beside it. Cheapest, and it is documentation of a measured fact.
• B. Say nothing about comments but mark the clause explicitly PR-body-only, so no one generalises it.
• C. Re-measure first. One comment on one thread is one data point; whether the rewrite is per-surface, per-tool (MCP add_issue_comment vs REST), or per-account is not established here. A second measurement through a different write path would settle it before anything is written down.

⛔ What I did NOT do

⛔ Did not edit AGENTS.md (out of scope for the card I was dispatched on, #10102). ⛔ Did not re-post or edit the report comment — the bare footer there is harmless and rewriting it would destroy the measurement. ⛔ Did not generalise from one sample: C is on the list for a reason.

Refs: #10102 (the card being reported when this surfaced) · PR #10204 · comment 5356851695

---

Generated by Claude Code

[qq9340100]

Triage first-touch (centralized round cont., session session_01C5fGRTUHXFevsEdMEFDadM): promote to pm:queue, domain:skills (AGENTS.md instruction-architecture), type Task, size S — option C then A: re-measure through a second write path first (one data point is not a rule; this triage session incidentally adds a second data point in the same direction — the facet-block HTML-comment marker was eaten on an issue comment at 15:59Z, #10255, confirming comment-channel rewriting is real), then scope the AGENTS.md clause to PR bodies and record the comment channel's measured behaviour beside it. Governed surface — draft PR, human merge.

---

Generated by Claude Code

[qq9340100]

Claim: PM loop round 7
Session: session_019T1sSZbQTnLhrK9HhNdNiB
Branch: claude/issue-10207-footer-clause-scope
Worktree: objectstack-issue-10207
Domain: domain:skills
File surface: AGENTS.md (the attribution-footer clause) ONLY — ⛔ no platform-readings row here (that file is owned by the #10360 dispatch this round) (stop on breach; explain in the report)
Container & model: S governed, mode:subagent, model: opus (AGENTS.md is governed but outside the narrowed clause ①'s protocol set; compensating control = this seat's fable review)
Clause-②: no
Serial constraints cleared: no open PR touches AGENTS.md. Ruled scope = C then A: re-measure the comment-channel rewrite through at least one additional write path before writing anything (one sample is one sample — the card's own caution), then scope the footer clause to what the measurements support, recording the comment-channel direction beside the PR-body rule.

---

Generated by Claude Code

[qq9340100]

Measurement probe M1 for this card (#10207, ruled scope C-then-A: re-measure before documenting). Write path: MCP add_issue_comment. Footer written in the session-URL form.

This comment exists only to be read back byte-for-byte; the table lands in the report comment.

---

Generated by Claude Code

---

Generated by Claude Code

[qq9340100]

Measurement probe M2 for this card (#10207, ruled scope C-then-A: re-measure before documenting). Write path: MCP add_issue_comment. Footer written in the bare form.

This comment exists only to be read back byte-for-byte; the table lands in the report comment.

---

Generated by Claude Code