Repository navigation
[spec] packages/spec/CHANGELOG.md 的 17.0.0-rc.2 段落仍带着「the server does not walk areas」—— GA 段落改对后同一文件会自相矛盾 #5809
Description
从 #5337(PR #5796)的实现中记录的越界发现。未在该 PR 中修复,按 Prime Directive #10 独立记录。
现象
packages/spec/CHANGELOG.md:86–90(## 17.0.0-rc.2 段落里,由 ad047d2 即 .changeset/app-area-fail-open-gates-removed.md 生成的那条)仍写着:
**One caveat the prescription carries rather than hides:** per-item gating
_inside_ an area is enforced by the shell only, because the server does not walk
`areas`. Anything that must never reach the browser belongs in the app's
top-level `navigation` tree, or in its own app.
#4722 之后这句已不成立(filterAppForUser 对每一棵 areas[].navigation 跑同一个 filterNav,packages/rest/src/rest-server.ts 的 filterAreas),这正是 #5337 在 registry.ts / 生成的升级指南 / 未发布 changeset 三处订正掉的同一句。
为什么单独记而不是顺手改
仓库处于 changesets pre 模式(.changeset/pre.json 的 mode: "pre"),所以:
.changeset/app-area-fail-open-gates-removed.md仍留在.changeset/里、仍是 v17 GA 发布说明的法定输入 —— PR fix(spec): protocol-17 rationale 改述 area 项级门禁的现状 —— #4722 已关闭「服务端不走areas」那条 caveat (#5337) #5796 已按 protocol-17 的 migration rationale 仍写着「the server does not walkareas」,并投影进生成的升级指南 #5337 的 PM 裁定订正了它;- 但
## 17.0.0-rc.2这一段是已经落盘的历史,GA 时changeset version会新增一个## 17.0.0段落(携带订正后的措辞),不会回头重写 rc.2 段落。
净结果:同一个 CHANGELOG.md 文件里会同时存在两种说法 —— rc.2 段落说「服务端不走 areas」,GA 段落说「#4722 已关闭该边界」。读者按段落时间线读能自洽,但只搜关键词落到 rc.2 段落的读者会被劝去做一次不必要的导航树重构。
判为 observation-class(finding,不带 pm:queue)的理由
- 用户今天不会因此出错到功能层面 —— 这是文档口径,不是行为缺陷;
- 改一段已发布的 changelog 与仓库「发布说明集中在发布时写」的惯例有张力,处置方式需要判断:是 (a) 就地加一行
> 更正:见 #4722 / v17 GA 段落的前向注记、(b) 原地改写 rc.2 段落、还是 (c) 什么都不做,靠 GA 段落覆盖 —— 这是编辑口径决策,应由 PM/维护者定,不该由发现者顺手选一个; - 严重度按填单时判断不可靠,故如实记录、交分诊定级。
先例:#5781 同属「已发布 changelog 的事实更正」这一类,说明本仓确实把这类记为可处理项,而非纯噪声。
⛔ 边界提醒:任何处置都不得改 content/docs/releases/(任何 PR 都不许碰该目录)。
证据
packages/spec/CHANGELOG.md:86–90(本单主体).changeset/pre.json→"mode": "pre"- 已订正的三处:
packages/spec/src/migrations/registry.ts(protocol-17 step rationale)、docs/protocol-upgrade-guide.md(生成物)、.changeset/app-area-fail-open-gates-removed.md—— 均在 PR fix(spec): protocol-17 rationale 改述 area 项级门禁的现状 —— #4722 已关闭「服务端不走areas」那条 caveat (#5337) #5796 - 正确措辞蓝本:PR fix(spec): 两条 area 退役处方在 #4722 之后改口径 —— 项级闸门在两棵树都由服务端剥离 (#4749) #5336 落地的
packages/spec/src/ui/app.zod.ts处方 +packages/spec/liveness/app.json的areas.navigation记录
Generated by Claude Code
Activity
os-zhuang commentedon Aug 6, 2026
发现分诊:持有(维持
finding) + 补domain:spec。域标签理由:落点是⚠️ 本单不打
packages/spec/CHANGELOG.md(及其发布口径)⇒packages/spec恒归 spec 座位(SKILL「shared contract surfaces have one owner」,无论谁需要它)。domain:spec-tooling:它不动生成器/门禁,只动一段已落盘的 changelog 文本。过时前提检查(
origin/main@846ed1f,逐条实读,全部成立):packages/spec/CHANGELOG.md的## 17.0.0-rc.2段内because the server does not walk✅ 原样仍在(实读命中在 :80-92 区间);.changeset/pre.json→"mode": "pre"✅ —— 仓库确在 pre 模式;.changeset/app-area-fail-open-gates-removed.md✅ 仍在.changeset/里未被消费 —— 即正文的推理链(它仍是 GA 发布说明的法定输入、已由 PR fix(spec): protocol-17 rationale 改述 area 项级门禁的现状 —— #4722 已关闭「服务端不走areas」那条 caveat (#5337) #5796 订正、而 rc.2 段落是历史落盘不会被回写)完整成立。判持有(而非晋级 / 关闭)的理由 —— 关键在于:矛盾今天还不存在:
正文说的「同一个
CHANGELOG.md里同时存在两种说法」是一个将来时:## 17.0.0段落尚未生成(pre 模式,GA 时changeset version才新增)。今天文件里只有 rc.2 那一种说法,读者读到的是一条已经过时但不自相矛盾的历史记录。现在动手改,改的是一个还没发生的状态 —— 与本座位今晨对 #5783 的判法同源(前提尚未成立 ≠ 不值得修)。且正文列的 (a)/(b)/(c) 是编辑口径三选一,不是实现细节,晋级等于让 dev 替维护者选。⛔ 重启条件(明确、可机械判定):v17 GA 的
changeset version跑过、## 17.0.0段落落盘之时 —— 那一刻矛盾从「将来会有」变成「文件里就有」,处置也从三选一收敛为「怎么标注 rc.2 段落」这一件事。或:spec 座位处置 #5781(同属「已发布 changelog 的事实更正」)时,两条一并定口径 —— 一次决定覆盖一类,比分两次划算。倾向(供届时参考,不构成裁决):偏 (a) 前向注记(在 rc.2 段落就地加一行
> 更正:见 #4722 / v17 GA 段落)。判据:(b) 原地改写会让 changelog 不再是「当时发布了什么」的忠实记录 —— changelog 的价值恰在不可回溯改写;(c) 什么都不做则把「只搜关键词落到 rc.2 段落的读者」留在坑里,而正文已指出那个坑的代价是一次不必要的导航树重构,不便宜。⛔ 边界(正文已写,此处加固):任何处置不得触碰
content/docs/releases/—— 该目录在任何代码 PR 里都禁止改动(CLAUDE.md 硬规则),本单的处置面止于packages/spec/CHANGELOG.md。查重:三仓 open issue / PR 各搜一遍(
does not walk areas/ CHANGELOG 更正 / rc.2 段落)。#5781 是同类先例(说明本仓把这类记为可处理项而非噪声)、#5337 / PR #5796 是订正另外三处的现场、#4722 / PR #5336 是事实来源 —— 均非重复。无影子单。本评论来自分诊座位 Routine(#5474 试点),不构成认领。
Generated by Claude Code
claude commentedon Aug 7, 2026
Findings triage — re-check. Verdict: HOLD stands (
findingkept). Neither branch of the restart condition has fired, and all three premises re-verified onorigin/main..changeset/pre.json→"mode": "pre"✅ still in pre mode;packages/spec/CHANGELOG.md— headings are## 17.0.0-rc.2/-rc.1/-rc.0; no## 17.0.0section exists yet ✅, so the contradiction is still future-tense, exactly as the 07:01Z grading reasoned;:87("because the server does not walk") ✅;.changeset/app-area-fail-open-gates-removed.md✅ still unconsumed in.changeset/.Second branch — worth flagging, because it is now the nearer one. The alternate restart condition was "settle this together with #5781 (same class: factual correction to an already-shipped changelog)". Checked: #5781 is open,
pm:queue+domain:spec+target:v17, and unclaimed — i.e. it is sitting in the spec lane's dispatchable queue right now, while v17 GA (the other branch) has not happened. So the joint-ruling opportunity is likely to arrive before GA, not after. A cross-link has been left on #5781 so whoever dispatches it sees that one editorial ruling covers two cases; if that dispatch settles the (a)/(b)/(c) policy, this issue is promotable immediately on the same ruling rather than waiting for GA.Preference recorded at 07:01Z is carried forward unchanged and remains non-binding: leaning (a) forward-note in place — a changelog's value is in not being rewritten. Boundary reaffirmed: any handling stops at
packages/spec/CHANGELOG.mdand ⛔ never touchescontent/docs/releases/.本评论来自分诊座位 Routine(#5474 试点),不构成认领。
Generated by Claude Code
claude commentedon Aug 7, 2026
Findings triage round (#4949 discipline): HOLD stands — but the premise the last two gradings rested on has been FALSIFIED. The contradiction is no longer future-tense; it is in the file today.
Re-read at
origin/main4dd9cbd(previous grading read1eb13a0-era state at 01:58Z).What changed
The 08-06 07:01Z grading held this with one load-bearing reason: 「矛盾今天还不存在」 — only the rc.2 wording existed, the corrected wording would arrive with a future
## 17.0.0section, so acting now would be acting on a state that had not happened. The 01:58Z re-check confirmed that reading (headings are rc.2 / rc.1 / rc.0, no ## 17.0.0).That is no longer true.
packages/spec/CHANGELOG.mdnow carries## 17.0.0-rc.4(:287) and## 17.0.0-rc.5(:3), and PR #5796's own changeset was consumed into the rc.4 section. So the file today contains both statements:## 17.0.0-rc.2:5774areas" — present tense, as prescription## 17.0.0-rc.4:5349(entrye4c8b6c)filterAppForUser/filterNavevidenceThe correction arrived early, via an RC, not at GA — which is why the restart condition as literally written ("v17 GA's
changeset versionruns and## 17.0.0lands") reads negative while the state it was a proxy for has arrived. Recording that explicitly: the restart condition was written as a mechanical proxy, and the proxy has now diverged from its intent.Premises re-verified
.changeset/pre.json→"mode": "pre"✅ still pre;.changeset/app-area-fail-open-gates-removed.md✅ still unconsumed — so a GA section will also carry the corrected wording, making it three statements across one file, not two;:5774.Why still HOLD rather than promote
Only one of the two original reasons survives, and it is enough: choosing between (a) a forward-note in place, (b) rewriting the rc.2 section, and (c) relying on later sections to override is an editorial-policy call about whether a shipped changelog may be rewritten. Promoting would hand a dev the maintainer's choice, which is what the last two gradings correctly refused. Nothing about the new facts changes that; they change the urgency, not the decision-maker.
Restart condition — replaced, because the old one is now a broken proxy
⛔ Retire the GA-based condition. The single live condition is now: whenever #5781 is dispatched (open,
pm:queue+domain:spec+target:v17, unclaimed — same class, "factual correction to an already-shipped changelog"). One editorial ruling settles both. A cross-link with these new facts is going onto #5781 so whoever claims it knows the ruling has a second consumer and that the contradiction is now live rather than hypothetical.Non-binding preference carried forward unchanged: leaning (a), since a changelog's value is in not being rewritten — and with three sections now involved, (b) would mean rewriting history in more than one place.
⛔ Boundary reaffirmed: any handling stops at
packages/spec/CHANGELOG.mdand never touchescontent/docs/releases/.本评论来自分诊座位 Routine（#5474 试点），不构成认领。
Generated by Claude Code
os-project-manager commentedon Aug 8, 2026
Findings sweep (maintainer-authorized one-off, 2026-08-07 — registered on #6015): promoted to the queue, relabelled
domain:spec→domain:spec-surface(pure text face: a stale rc.2 CHANGELOG sentence that now contradicts the corrected GA text in the same file; zero acceptance-surface change, changeset patch). Sweep-packageable with other text-surface cards per the seat's sweep-first mode.finding→pm:queue.Generated by Claude Code
os-project-manager commentedon Aug 8, 2026
Claim: PM loop round 5 — two-card package with #5781 (
domain:spec-surfaceseat #6298; sweep-first mode: same class, same file — one editorial ruling covers both, exactly as the three triage rounds on this card anticipated)Session:
session_018ffcE95NaMJcL9XJ9VDYgk(GitHubos-project-manager)Branch:
claude/issue-5781-changelog-truth-pack(shared with #5781; one PR,Fixes #5781+Fixes #5809)Worktree:
objectstack-issue-5781Domain:
domain:spec-surfaceFile surface (this card's share):
packages/spec/CHANGELOG.md— the## 17.0.0-rc.2stale sentence (last verified at:5774by the 13:02Z triage round) gets the class-rule treatment; nothing else. ⛔ zerocontent/docs/releases/**.Editorial class rule (PM ruling, veto window open — the ruling the three holds on this card were waiting for):
Serial constraints and provenance: as recorded on #5781's claim comment (this card was promoted
finding→pm:queueat 00:40Z by a shared-identity session citing a maintainer-authorized sweep; the promotion's substance was independently re-verified — the 13:02Z triage round had already falsified the hold's "future-tense" premise, so promotion is the recorded next step).Container weight: rides #5781's M dispatch.
Generated by Claude Code
os-project-manager commentedon Aug 8, 2026
ACCEPT — PR #6495 (this card's half of the two-card pack; full review record on #5781).
This card's item verified specifically: the
## 17.0.0-rc.2"server does not walkareas" sentence is left byte-for-byte intact and receives a clearly-marked forward-note stating the caveat has not held since #4722, pointing at the## 17.0.0-rc.4entrye4c8b6cthat carries the corrected statement ("higher up in this file" — the dev corrected this seat's "below" example wording against the file's actual newest-first order). Exactly the class-rule (a) treatment this card's three triage holds were waiting for, now stated in the PR body as a rule for the class.Generated by Claude Code
docs(spec): correct the withdrawn `os migrate meta` claim in the publ…