docs(pm): rest-channel.md rows 1–2 name the ccr REST routes for the draft flip and auto-merge

[claude]

The session proxy refuses POST /graphql outright and its refusal names first-party REST routes on the seat's own credential, so rest-channel.md rows 1 and 2 — the draft → ready flip and auto-merge / enqueue — are no longer GraphQL-only with no fallback. Both routes were driven with read-back on 2026-09-12, by the cli seat at 07:46Z and by the skills seat at 08:45Z. The two rows therefore leave the 「不可迁移」 bucket for the write-side list as ccr bullets that keep MCP as the alternative rather than the only channel, keep the read-back criteria unchanged, and record in one clause the cost the new channel brings: the timeline actor differs by channel.

Fixes #17820

Acceptance notes

Premises, checked on origin/main before the first edit

1. ✅ Rows 1 and 2 read exactly as the card quotes them. The file is byte-identical to the dispatch's 2070a1d8 (git diff 2070a1d8 HEAD -- .claude/skills/pm-dispatch/references/rest-channel.md is empty); the rows sit at lines 53–57 on c1078a559, not 51–56, which is a line-number drift only.
2. ⚠️ FALSE in its first half, and it is load-bearing. The dispatch and the grading comment both say the file is "82 lines, not in the line-count ratchet's CEILINGS table". It IS in that table — ['.claude/skills/pm-dispatch/references/rest-channel.md', 82] in scripts/pm/check-skill-line-ratchet.mjs, i.e. the file sits at headroom 0, and it is additionally recorded as a SOURCE (93 → 82) inside a declared CROSS_FILE_MOVE, so its ceiling number is load-bearing arithmetic for another file's raise. The second half of the premise holds: MAX_LINE_BYTES = 120 applies per line, and every line written here is within it (the widest is exactly 120 bytes).
3. ✅ No open PR touches rest-channel.md — checked by reading the file list of all 20 open PRs, not from memory.

How the required content was paid for, at net zero lines

The rewrite needed seven bullets where the two retired rows occupied five lines, and a ratchet at headroom 0 admits no raise without a maintainer ruling. The missing two lines came from restoring an encoding a standing maintainer ruling already prescribed for this very file, quoted in the ratchet script's own CEILINGS comment (2026-09-01, 总监批): 「C 同批:退役 L58 对红窗常设规则的复述 —— 内容归属裁定:红窗规则由 platform-readings.md 配额段独家持有,rest-channel.md 只留指路」, which the same comment records as having been paid "folded into the 不可迁移 heading at zero line cost". The pointer had since drifted back out into a standalone line plus its blank; it is folded back into the heading here, verbatim in substance. Result: 82 lines before, 82 lines after, ceiling 82, headroom 0 — the file's own convention, and the CEILINGS entry is untouched.

What landed, per work item

• A — POST .../pulls/{n}/ccr/ready_for_review and the reverse .../ccr/convert_to_draft are named as the seat-credential channel; MCP update_pull_request stays as the alternative; the read-back criterion 「状态码不作数,读回才作数」 stays, now spelled with both halves (GET /pulls/{n} → draft: false, timeline ready_for_review). The 「断粮:等它恢复或人工点一下」 clause is gone.
• B — PUT .../pulls/{n}/ccr/auto_merge with {"merge_method":"SQUASH"}, DELETE disarms; MCP enable_pr_auto_merge stays as the alternative; enqueue is verified by added_to_merge_queue on the timeline and the landing shape by git rev-list --parents, with 「⛔ auto_merge 与回显都不作数」 keeping both unreliable readings out. 「配额红窗无退路」 is gone; the direct-merge note PUT .../pulls/{n}/merge is kept.
• C — one clause, on the row that carries the direct-merge note: 「ccr 的 timeline actor 记 claude[bot],MCP 记席位账号」. Stated as a fact, with no recommendation and no ruling; the card number is deliberately absent because pnpm check:pm-skill-id-lint bans issue IDs in this corpus, so the question's owner cannot be cited in-file.
• D — both rows moved into 「写侧 —— 全部可迁移」, beside the other PR-lifecycle writes. The 「不可迁移」 framing is re-counted honestly: 「只有这几件」 → 「只有这三件」, and rows 3–5 renumber to 1–3 with their text untouched.
• E — both new rows carry 「2026-09-12 两席实调」; the retained 2026-09-11 reading keeps its own date.

Reading kept rather than dropped

The old row 1's naked-PATCH reading (「PATCH /pulls/{n} 带 {"draft": false} 回 200 且什么都没改」) is kept, as the ⛔ line directly under the new route. It is what stops a seat from reaching for the plain REST route now that a REST route exists, and it is the evidence behind 「状态码不作数」.

Verification

node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack in this worktree, derived at d6ebc6383, printed 14 commands; all 14 were run and all exited 0. Verdict lines worth quoting:

✓ check-skill-line-ratchet: .claude/skills/pm-dispatch/references/rest-channel.md is 82 lines (ceiling 82; headroom 0).
✓ check-skill-line-ratchet: .claude/skills/pm-dispatch/references/rest-channel.md: widest table row is 0 bytes (pin 0; headroom 0).

Two of the fourteen first exited 3 — PREREQUISITE NOT MET, not a finding: pnpm install had not run in this fresh worktree, and check:doc-formula-expressions additionally needed @objectstack/formula and @objectstack/lint built. Both were re-run green after pnpm install and a lock-held turbo run build --filter=@objectstack/formula --filter=@objectstack/lint. No changeset: the diff is .claude/** only, which publishes nothing from any released package, so the PR carries skip-changeset.

Out of scope, reported not done

references/platform-readings.md now contradicts this table in three of its lines — 「undraft 可用路径只有 MCP 这条」, 「裸 GraphQL 在会话内被拒,它建议的改用 REST 对 undraft 是错的」 (the suggested route now works) and the enable_pr_auto_merge rows that read as the only enqueue channel. The card explicitly rules that file out of this landing, so nothing here touches it; the skills lane owns the follow-up. Also noted, not filed: the actor question — an act through a ccr route is written to the timeline as claude[bot] and through MCP as the seat account — belongs to the open card that holds it, and this PR rules nothing about it.

维护者速读(草稿)

改了什么 — PM 用的 REST 通道对照表里,「draft 转 ready」和「挂 auto-merge / 入队」两条原本写着「只能走 GraphQL、配额红窗里无路可走」。今天平台不再这样:会话代理直接拒绝 GraphQL,并在拒绝里点名一组席位凭据可用的 REST 路由。这两条改写成「有 REST 通道、MCP 仍是备选」,并从「不可迁移」挪到「写侧」,同时保留原有的「读回才作数」判据。

为什么改 — 表里的事实错了,代价不是掉链子而是排错计划:席位按旧表会在配额红窗里干等或去找人工点按钮,而它自己的通道是通的。两次实调(2026-09-12,cli 席 07:46Z、skills 席 08:45Z)都带读回验证。

风险与代价(含回滚) — 风险有一条并已写进表里:走 ccr 路由时,时间线上的操作者记成 claude[bot],走 MCP 记成席位账号;哪个对是另一张卡的问题,本 PR 不裁。回滚成本为零 —— 单文件文档改动,git revert 即可,无代码、无发布物、无 changeset。行数守恒:82 行进、82 行出,棘轮天花板与其跨文件移动声明都没动。

席位意见 —

你要做的 — 这是受管面(.claude/**),按规矩只能由维护者人工合并;PR 停在 draft,席位不会翻 ready、不会挂 auto-merge。需要你确认的只有一点:本 PR 把「红窗规则只在 platform-readings.md、本表只留指路」那条既有裁决的编码方式(折进「不可迁移」标题、零行成本)恢复了原样,以此腾出两行装下新内容。若你希望改用「把天花板从 82 提到 84」的方式承载,请直接说,届时需要你的一句提额裁决。

Generated by Claude Code

---

Generated by Claude Code

[claude]

维护者速读

改了什么 — PM 的 REST 通道对照表(references/rest-channel.md)里,「draft 转 ready」与「挂 auto-merge / 入队」两条原本写在「不可迁移」段:只能走 GraphQL(即 MCP),配额红窗里无路可走、只能等或找人点按钮。今天平台不是这样:会话代理直接拒绝 GraphQL,拒绝信息里点名了一组席位自己凭据就能用的 REST 路由(…/pulls/{n}/ccr/ready_for_review、…/ccr/auto_merge 等)。两条改写成「有 REST 通道,MCP 是备选」,挪进「写侧」段,原有的「状态码不作数、读回才作数」判据原样保留。

为什么改 — 表里的事实错了。代价不是掉链子,是排错计划:席位按旧表会在 MCP 限流时干等,而它自己的通道是通的。今天两次实调都带读回:cli 席 07:46Z(PR #17812),本席 08:45Z(PR #17816,MCP 刚被限流,改走 ccr 路由,入队事件落在时间线上)。

风险与代价(含回滚) — 一条已写进表里、不裁:走 ccr 路由时时间线上的操作者记成 claude[bot],走 MCP 记成席位账号;哪个对是 #11742 的问题。回滚零成本:单文件文档改动,git revert 即可,无代码、无发布物、无 changeset。行数守恒 82 → 82;文件在棘轮里天花板 82、余量 0,dev 用恢复既有裁决的编码方式(把「红窗规则住 platform-readings」那条指路折回段标题,2026-09-01 总监批 C 条原本就这么定的,后来漂成了独立行)腾出两行,天花板和跨文件移动记录都没动。

席位意见 — 接受(ACCEPT 5644958470,契约档复核)。附带更正一处本席自己的误读:派发时说这文件「不在棘轮里」,实际在,余量为 0;dev 的净零行支付是对的做法。

你要做的 — 受管面(.claude/**),只能人工合并:确认 PR #17823 的 11 行改动后合并即可;若你更希望用「天花板 82 → 84」来承载而不是折回标题,说一句,本席重派。

---

Generated by Claude Code