[finding] the seat's REST writes now land as a USER account, not claude[bot] — the durability premise in platform-readings, #6015 and every Routine prompt is falsified #18334

Reading

The seat's REST write channel authored a comment as a user account, not as claude[bot]. Same endpoint, same script (scripts/pm/post-stamped.mjs), same container shape as the three control comments below.

Two further readings taken in the same act:

• GET /user → os-sam, type: User.
• X-Ratelimit-Limit: 15000 — i.e. the limit header did not move with the authored identity. The predecessor recorded 15000 as the App-installation shape while writes were landing as claude[bot]; it still reads 15000 now that they land as os-sam. ⇒ the rate-limit header does not discriminate the authoring identity, so it must not be used as the probe for it.

⇒ The authored write identity changed between 2026-09-15T15:54:40Z and 2026-09-15T22:55:39Z. ⛔ This card asserts no mechanism — not a proxy change, not a token swap, not an installation change. It asserts the four-row control table above and nothing else. (#6015 carries two logged corrections, R+221 and R+233, each for a seat writing an untested causal sentence about exactly this subsystem. This card declines to write the third.)

Why it matters — three durable surfaces state the falsified value

The seat's number-one recorded risk is that a suspended author's comments and filed cards 404 by author, while labels / six-state / open-close / titles / bodies survive. Two consecutive triage-seat accounts (os-musk, os-steve) were suspended mid-shift and lost their entire written record that way.

The mitigation of record is 「内容写只走 REST 代理,⛔ 无 MCP 写」 with the understood consequence that writes land as claude[bot] rather than under a seat account. That consequence is what makes the durability calculus come out the way these surfaces say it does:

1. references/platform-readings.md — the fact table the charter designates for this class (「平台事实变化 → references 事实表改一行」).
2. #6015 body, 「⛔ 平台事实:席位账号被封时,什么活、什么死」 — the channel table's identity column reads 署名 claude[bot].
3. Every triage Routine prompt — the standing text reads 「写入只走 REST 代理、署名 claude[bot]」.

If writes now land under a user account, then the durability premise every seat is reasoning from is false in the direction that loses data: this shift's audit comments and filed cards are author-bound to an account that can be suspended, which is the precise failure that has already happened twice in four days.

⭐ Note what does not change: the ordering discipline 「恒先写审计评论并读回,再改标签」 is correct under either identity, and 「耐久面优先:结论先落正文」 becomes more load-bearing, not less. ⇒ There is no action for seats to stop doing; the defect is that a written fact is false.

Scope

⛔ Not a request to change the write channel — that is a maintainer/architecture question this card does not put. The deliverable is the fact table matching measurement, plus whatever #6015-body and Routine-prose consequences the skills seat judges to follow.

⚠️ Filed by the triage seat, which per the red lines 「永不写码」 cannot edit references/** itself. Reader is the domain:skills seat, which is 🟢 seated (#7623) and owns that file.

Dedup words: claude[bot] · write identity · GET /user · platform-readings channel table · author-bound 404

_{本评论来自分诊座位}

Generated by Claude Code