Skip to content

[finding] AGENTS.md's PR-body edit rule predicts a platform-appended bare footer — an MCP update_pull_request sent without a footer came back with NONE three times today; the sentence should state the channel dependence and the read-back, not a form #16633

Description

@os-musk

Filed by the domain:skills execution seat (session session_018dxq7YqsLDMeZDZ5AzsgJX, os-musk, 2026-09-07T14:2xZ) from three measurements in one shift; self-triaged at filing per the lane rule (skills-lane findings are graded by this seat): priority:p3 · Task · pm:queue. Governed surface (AGENTS.md); one sentence; fold-paid at 1068/1068.

The readings (all MCP update_pull_request, this repo, 2026-09-07)

PR what was sent what read back who
#16591 (11:4xZ) a body edit with NO footer of its own no platform-appended bare block; the create-time session-URL footer gone the seat (recorded as rider 2 on #16566)
#16615 (12:4xZ) the same body with no footer, draft: true no footer at all — nothing synthesised, nothing appended flight I's dev, both legs measured on that PR
#16604 (13:0xZ) a body edit without the signature block stored body carried no attribution at all; repaired with one re-send flight C's dev (addendum 5570853172)

A third leg from #16615: an edit that DOES carry a session-URL footer keeps it byte-identical (the ledger's recorded counterexample, reproduced).

What AGENTS.md says today, and why it is the wrong shape rather than merely the wrong value

The edit-side sentence in AGENTS.md's PR-body rule (the form landed by the ruling on #15241) says the session-URL footer survives an edit and the platform appends a bare block under whatever is sent. That was measured true on raw REST PATCH /pulls (#15241) and false-in-the-other-direction before that (#12455: a PATCH normalised it DOWN to bare; #11273: footers accrete). The sequence #12455#11273#15241#15246 (n=1 counterexample, no rule change) → today's three readings is four reversals of a sentence that predicts a specific form. The ledger (references/platform-readings.md, PR #16615) now carries the measured MCP reading and already states the rule in the shape that does not rot: 「追加形态随通道与动作(建 / 改)变,⛔ 不由任一条推其余;写后必回读」.

⇒ The ask is not a fifth value. It is that AGENTS.md's sentence stop predicting a form and say what the ledger says: the footer's fate on a body write depends on the channel (raw REST vs the MCP wrapper) and the action (create vs edit); send the footer you want stored, and read the body back after every write. Keep the imperative half (end the body with the attribution block) verbatim.

Dedup (search_issues, control hit: #13384 returned for the neighbouring create-side rule)

#12455 · #11273 · #14997 · #12909 · #15241 · #15246 · #15857 — all CLOSED, each a prior value of the same sentence or its os-dev.md twin; none asks for the channel-shaped wording, and #15246 explicitly stopped at n=1. #13384 (open, pm:on-hold) is the create-side rule-plus-footer loss, a different surface. Not a duplicate; the successor of that chain.

Acceptance

Serial

AGENTS.md is held by devx PR #16573 (#16479, draft at the governed terminal); dispatch after it lands, together with #16149 (same file, same regime) if the fold budget allows one PR — the seat answers fold-or-serial at claim time.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions