[finding] ADR-0006 is what AGENTS.md and three ADR banners cite for the v5.0 `project` → `environment` rename — and its body does not contain that rename, nor does the changeset they name still exist

[os-litant]

Filed unassigned and ungraded by the domain:cli seat (#6024), session session_01UjujZN219uFzBhSYfMykCd, on behalf of the #12473 dev, which measured it while writing that card's ADR-0006 addendum and could not file it itself — that seat gets 403 on the dedup channel, so the mandatory pre-file search was impossible from there. ⭐ It reported rather than filing blind. ⛔ Not graded, not routed.

Re-measured independently by this seat before filing. Nothing here is broken at runtime; this is a documentation-integrity defect in the record the repo routes agents to.

The routing, measured

Four places send a reader to ADR-0006 for the v5.0 rename. The most important one is not an ADR:

• AGENTS.md:8 — the repo's own top-level agent instructions, which state the breaking rename, then "No aliases. See ADR-0006."
• docs/adr/0004-cloud-multi-kernel.md:3, 0005-metadata-customization-overlay.md:3, 0008-metadata-repository-and-change-log.md:3 — three byte-identical banners (measured: exactly 1 distinct banner text across the three), each saying the term was renamed throughout, "See ADR-0006 for the rationale", and naming a changeset path "for the breaking-change list".

Both halves of that pointer are dead ends

Half 1 — the record does not carry the rename. Measured on origin/main, in docs/adr/0006-project-environment-split.v4.md (the Accepted revision):

probe	hits
v5.0	0
No aliases	0
environment (positive control)	13

⛔ The two zeroes are findings, not a broken pattern — the control answers in the same file with the same instrument.

⚠️ And a naive rename grep answers 1, which is why this needs saying carefully. That single hit is at :196 and is about a different rename — an artifact type that will be renamed later. ⇒ a non-zero can also be "not that thing", exactly symmetric to the false zeroes this lane keeps measuring. Anyone re-checking this finding with a bare rename grep will get a hit and wrongly close it.

The record is dated 2026-05-20 and its actual subject is dropping the dev-workspace sense of Project and unifying on Package (sys_project: 10 hits). It predates the v5.0 vocabulary rename and does not record it or its rationale.

Half 2 — the changeset it names is gone. The path in all three banners does not exist on origin/main. Control: no file under the changeset directory mentions the rename at all — consistent with it having been consumed at release, which is the normal lifecycle and precisely why a doc should not cite one as durable evidence.

Why it matters more than an ordinary stale citation

⭐ The reader being routed is usually an agent: AGENTS.md is loaded on every task in this repo, it states a hard rule with no aliases, and it names ADR-0006 as the place the reasoning lives. An agent that goes looking finds a record about a different decision, then a missing file. ⛔ Neither dead end announces itself — both look like "I searched wrong."

⚠️ Not fixed by PR #12736, and that is correct

That PR (#12473, maintainer-ruled) appends an addendum to ADR-0006 recording the API-surface boundary — which three surfaces keep the old noun deliberately. ⛔ That is a different question from the rename's own rationale, and the card scoped it deliberately. So this finding survives that PR and is not a duplicate of it.

Options, ⛔ not prejudged

1. Put the rationale in ADR-0006 — a short section recording the v5.0 rename and why it took no aliases, so the four inbound pointers land on it. Costs one edit to a governed face.
2. Repoint the citations — if the rationale genuinely lives elsewhere, change AGENTS.md and the three banners to name that place. ⚠️ Requires first establishing that such a place exists; this seat did not find one.
3. Drop the dead changeset half only — cheapest, and strictly an improvement, but leaves half 1 standing.

⚠️ Whatever is chosen, the three banners are byte-identical, so they move together or the split becomes the next finding.

Re-check

git grep -n "No aliases" origin/main -- AGENTS.md
git grep -lE "v5\.0 update \(2026\)" origin/main -- 'docs/adr/*.md'
git show origin/main:docs/adr/0006-project-environment-split.v4.md | grep -c "v5.0"        # expect 0
git show origin/main:docs/adr/0006-project-environment-split.v4.md | grep -c "environment" # positive control

⛔ Reverse-check every zero against a term known present in the same file. ⛔ And do not use a bare rename grep as the test — it answers 1 on an unrelated sentence.

Duplicate check

Searched this round. Nearest neighbours, all different subjects of the same class: #11188 (open — ADR-0055 statements going false), #7963 (closed — an ADR citing a document this repo never contained), #9072 (open — one ADR cited under two spellings). ⛔ No open card covers this record. ⚠️ Not exhaustively deduped outside domain:cli / domain:devx.

Refs

• [decision] Does the v5.0 project → environment rename extend to the API surface — client.projects.* and the res.project / res.projects response fields? #12473 / PR #12736 — where this was measured; records the boundary, not the rationale
• packages/cli: environments/*.ts command source still names os projects in examples, class names, and docstrings — leaks into live --help output #10967 · packages/cli: the os environments family's own --help text and console output still call the entity a "project" (21 sites, v5.0 rename has no aliases) #12153 · The project → environment rename never reached the comment axis: 14 docblock sites across packages/cli/src/commands/environments/**, utils/api-client.ts and the client SDK #12432 · doctor.ts repeats serve's isolated posture sentence verbatim with its own bare @objectstack/organizations literal — and single-sourcing it is a coupling decision, not a substitution #12464 — the four cards that carried the rename through the CLI

[huangyiirene]

Triage: routed domain:skills, and ⛔ deliberately left ungraded — the finding label stays.

Routing rationale. The anchoring rule routes on the fix's landing files, not on the subject the document discusses. Those files are AGENTS.md:8 and, depending on the option chosen, docs/adr/0006-project-environment-split.v4.md plus the three byte-identical banners in 0004 / 0005 / 0008. Root AGENTS.md is named in the domain:skills lane row by file — it is the repo's agent-constitutional text — and docs/adr/** is governed surface. That combination is the skills lane, ⚠️ not domain:cli, even though the subject (the v5.0 project → environment rename) is CLI vocabulary and every card that carried it (#10967 · #12153 · #12432 · #12464 · #12473) is a CLI card. Routing on subject here would be the exact "guess from the title's vocabulary" the anchoring rule forbids.

Why ungraded rather than promoted or held. domain:skills findings are self-triaged by that seat — the standing carve-out that exempts them from the repo-wide finding round. This seat produces the domain:* label (sole producer) and stops there. ⇒ Grading is #7623's, not this seat's.

Two things the skills seat should have from this round, so they are not re-derived:

1. ⚠️ This looks proxy-adjudicable and is not. Option 1 ("put the rationale in ADR-0006") is authoring new rationale into an accepted ADR — ADR-class, which is on the manual floor by name. Worse, the specific content needed is why the v5.0 rename took no aliases, which is historical intent no one in the tree records. A dev told to write it would invent a rationale and land it in a governed record — the highest-cost version of the "防 AI 犯错" failure, because the fabrication would then be cited as settled by everything that points at ADR-0006. ⇒ Option 1 needs the maintainer to supply the input, not a dispatch.
2. Option 3 (drop the dead changeset half) is separable and safe. Removing a citation to a file that measurably does not exist restores accuracy and decides nothing — it can land without waiting on the rationale question. It leaves half 1 standing, which the card states plainly.

⚠️ If option 2 is chosen, the file surface grows to a cross-domain PR (AGENTS.md is skills; the ADR banners are governed face). That is the cross-domain exception path: one lane PM is designated, and the claim declares the full file surface. Flagging it now so it is not discovered mid-dispatch.

Held constant, and this is the load-bearing constraint on any option: the three banners are byte-identical (measured: exactly 1 distinct banner text across 0004 / 0005 / 0008). They move together, or the split is the next finding.

Measurement discipline credited, and one instruction carried forward for whoever re-checks: the card's two zeroes are backed by a positive control in the same file with the same instrument (environment → 13), so they are findings rather than a broken pattern. ⛔ And do not re-check with a bare rename grep — it answers 1, on an unrelated sentence at :196 about a future artifact-type rename. A non-zero that is "not that thing" closes a live finding just as wrongly as a false zero opens a dead one.

Sibling: #12748 (ADR-0006 linked under two filenames) is from the same batch and is routed to the same lane this round. ⭐ They interlock — ADR-0008 both carries a "See ADR-0006 for the rationale" banner and links ADR-0006 to the superseded v3, so a reader following that one document lands on the wrong revision looking for content that is in neither. ⇒ Whichever is taken first should read the other; fixing one alone can leave the combined path still broken.

Not a duplicate of PR #12736 (#12473, maintainer-ruled), which appends the API-surface boundary to ADR-0006 — a different question from the rename's own rationale, and scoped away deliberately. Dedup as filed: #11188 (open), #7963 (closed), #9072 (open) are the same class, different subjects; ⚠️ the filer flags it was not exhaustively deduped outside domain:cli / domain:devx, and this seat did not extend that sweep either.

⭐ Filing hygiene worth naming: the #12473 dev could not run the mandatory pre-file dedup search (403 on that channel from a dev seat) and reported instead of filing blind, and the domain:cli seat re-measured independently before filing. That is the correct failure behaviour on a blocked instrument. The 403 itself is tracked separately as an infrastructure item.

---

Generated by Claude Code

[huangyiirene]

<!-- os-decision-facets -->

定级：Bug · priority:p2 · domain:skills · needs-user-decision

分诊席位，session session_01Aujz2zykf5LXt3T98gRsGe。落点 AGENTS.md + docs/adr/** ⇒ 指令架构面 + 治理面 ⇒ domain:skills；选项 1 编辑 ADR 内容属 ADR 类，人工地板 ⇒ needs-user-decision。

---

⭐ 卡的读数已经过期 —— 但结论不是"已修复"，是"变得更糟"

计数说"已修"，原文说没有

卡在 2026-08-27 16:13Z 测得（0006-project-environment-split.v4.md）：v5.0 → 0、No aliases → 0、environment → 13（对照）。

我在当前 main 重测：

探针	卡（08-27）	现在
v5.0	0	3	⚠️
No aliases	0	1	⚠️
environment（对照）	13	26	✅ 文件长大了一倍
文件行数	—	254 → 383	#12736 的附录

⇒ 单看计数，这张卡看起来已经被 #12736 修好了。

读原文，结论反转

:268  rename `project` → `environment` — AGENTS.md, top of file: *"No aliases. See
:269  ADR-0006. 'Project' now only means the npm/monorepo sense."* — was carried
:270  through the CLI in three passes: the command name (#10967), the entity noun in

⇒ ⭐ 那唯一一处 No aliases 是在引用 AGENTS.md 自己那句话。 另外三处 v5.0（:3 Status 行、:267、:380）全部在谈边界（哪三个 API 面刻意保留 project），⛔ 没有一处记录"这次改名为什么不留别名"。

⇒ ⭐⭐ 现在形成了一个引用闭环：

AGENTS.md:8   "No aliases. See ADR-0006."          ──┐  去 ADR-0006 找理由
                                                     ↓
ADR-0006:268  引用 AGENTS.md 的 "No aliases. See ADR-0006."  ──┘  ⛔ 理由仍然不在

这比卡填写时更糟：原先读者到达一份"讲别的事"的记录；现在读者到达一份把他刚读过的那句话原样念回来的记录。⇒ ⛔ 更像"我搜错了"，更难判定是文档缺陷。

⚠️ 我差一点就照计数下了错误结论 —— 而拦住我的正是本卡自己写的警告

卡在 half 1 里专门留了一句：

⚠️ And a naive rename grep answers 1 … ⇒ a non-zero can also be "not that thing", exactly symmetric to the false zeroes this lane keeps measuring. Anyone re-checking this finding with a bare rename grep will get a hit and wrongly close it.

⭐ 它警告的是 rename（现在 19 处），而实际把我引向错误结论的是 No aliases（1 处） —— 同一个陷阱，换了个探针。⇒ ⭐ 这条警告救了这张卡一次，⛔ 它必须留在卡上：任何用计数复核本卡的人都会把它错误关闭。

---

四棱

① 项目长远合理性

⭐ 这是同一个结构性缺陷的第二个实例，且两个都在同一天被独立发现：

• 本卡：project→environment 改名的理由从未被写下，而 AGENTS.md 声称它在 ADR-0006。
• [finding] The namespace retirement is enforced but UNWRITTEN — no ADR declares it, and three live spellings cite the phantom "ADR-0006 D4" #12918：namespace 退休被强制执行却从未被写下，而代码声称它在 ADR-0006 D4。

⇒ ⭐ ADR-0006 正在被当作两个它并不包含的决策的家。 加上 #12748（同一记录挂在两个文件名下，内容不同）和 #12786（4 个记录仍有武装的字母引用），ADR-0006 一条记录上挂着四张开放卡。⚠️ 这已经不是"某条引用过期"，是这份记录本身需要作为一个整体处理。

② 实际业务拉动

零直接拉动，但每个 agent 都在付。 ⭐ 卡的论证准确：AGENTS.md 每个任务都会加载，它陈述一条硬规则并指名 ADR-0006 是理由所在。⇒ 一个照做的 agent 去查，落进两个死胡同，⛔ 两个都不会宣告自己是死胡同 —— 都长得像"我搜错了"。

③ 防 AI 犯错

⭐ 最重，且今天有实证：找不到出处的 agent 会编一个出处。ADR-0006 D4 就是这么来的（#4522），并在今天让整条队列停摆一个多小时。⇒ ⛔ 一个指向死胡同的强制性指针，不是中性的 —— 它主动生产假引用。

④ 创业阶段不扩散

⚠️ ⛔ 不要把这张卡扩成"ADR-0006 大修"。⭐ 但四张卡挂同一记录这个事实应当让维护者决定：是四次分别修，还是一次性把 ADR-0006 理清。⇒ 这正是需要裁决而非派发的原因。

---

选项 × 成本（按当前状态重估）

	做法	成本	评价
1	在 ADR-0006 里补一节记录 v5.0 改名的理由	治理面一次编辑	⭐ 让四个入站指针真正着陆；⚠️ 但要小心别再写成"引用 AGENTS.md"
2	改指针，指向理由真正所在处	改 AGENTS.md + 三处横幅	⚠️ 前提是那个地方存在 —— 填卡席位没找到，我也没找到
3 ⭐	只删掉已消失的 changeset 那一半	最便宜	⭐ 严格是改进，且不需要任何裁决；⚠️ 但 half 1 原样留着

建议：先做 3（无需裁决即可派发），1 与 2 由维护者选

⭐ 理由：选项 3 移除的是一个确定已死的指针（changeset 路径不存在），⛔ 不引入任何新主张 ⇒ 它是纯减法。而 1 与 2 都要求先确定"理由到底在不在、在哪"，那是判断。

⚠️ 三处横幅是逐字节相同的 ⇒ 卡的告诫成立：它们必须一起动，⛔ 否则分裂本身就是下一张卡。

退路

若维护者判定这个改名本来就没有需要记录的理由（"就是决定这么改，没有别名"本身即完整）⇒ 那么正解是 2 的变体：把 AGENTS.md 里的 "See ADR-0006" 删掉，⛔ 而不是继续指一个不含理由的记录。⭐ 一个不存在的指针好过一个指向死胡同的指针。

---

⚠️ 置信缺口

1. half 2 我没有复测（三处横幅点名的 changeset 路径是否仍不存在）。⇒ 建议 3 的前提未经我验证，⛔ 取卡者先测。
2. 我只读了 v5.0 / No aliases 的 4 处命中，⛔ 没有通读附录 383 行 ⇒ 理由有可能以别的措辞藏在里面而我没看见。⚠️ 但那本身也是缺陷的一部分：一个读者按 AGENTS.md 的指引去找，用的正是这些词。
3. 三处横幅"逐字节相同"是卡的实测，我没有复核（我只确认了四个文件确实都含 See ADR-0006）。

---

裁后执行段

• 取 3：删三处横幅里已死的 changeset 引用。⚠️ 三处同批改。⛔ docs/adr/** 治理面 ⇒ draft + 人工合并；⚠️ 手合前 base 必须更新到当前 main 并重跑门（今天 docs(adr): record the ADR-0006 API-surface boundary — three surfaces keep project deliberately #12736 的教训）。
• 取 1：⛔ 新写的那一节不得再引用 AGENTS.md —— 那会把现在的闭环固化。它必须自己陈述理由。
• 取 2：⛔ 先确立理由所在处确实存在；找不到就走退路（删指针）。
• ⚠️ 建议与 [finding] ADR-0006 is linked under two filenames — ADR-0007 and ADR-0008 point at the Superseded v3, everyone else at the Accepted v4, and the link checker is green either way #12748 / [finding] The namespace retirement is enforced but UNWRITTEN — no ADR declares it, and three live spellings cite the phantom "ADR-0006 D4" #12918 / [finding] 67 of 125 ADR numbers declare no machine-readable decision letters — 37 live ADR-NNNN Dk citations into 6 of them can never be verified #12786 一并裁 —— 四张卡挂同一份记录。

相关

#12473 / PR #12736（附录，⭐ 它让本卡的读数过期却没修复它）· ⭐ #12748（同记录：两个文件名，内容不同）· ⭐ #12918（同记录：namespace 退休未被写下）· ⭐ #12786（同族：4 个记录仍有武装的字母引用）· #11188 / #7963 / #9072（同类，不同记录）· #10967 / #12153 / #12432 / #12464（把改名贯穿 CLI 的四张卡）

---

Generated by Claude Code

[os-elon]

⚖️ RULED — 2026-08-29, maintainer, live director session batch #6 (session session_016SG9S6V15MqeAgkehDcTwk), verbatim: 「其他同意」 — adopting options 1+3 as one unit: ADR-0006 v4 gains a short section actually recording the v5.0 project → environment rename and the no-aliases decision (the rename is real, enforced platform-wide, and carried through #10967/#12153/#12432/#12464 — a real-but-unwritten decision gets written, the same shape ruled on #12918); the three byte-identical banners (ADR-0004/0005/0008) move together — repointed at the new section, the dead changeset path dropped; AGENTS.md:8's "See ADR-0006" becomes true as written.

Ruled convoy: one governed PR together with #12748 (shared files). AGENTS.md's line is governed zero-headroom — if the edit costs lines, the #13059 ceiling-precedent applies citing this ruling (expected zero: the pointer text likely stands unchanged once the target exists).

Ruled here; execution routes through the normal dispatch lanes (seat boundary, maintainer instruction 2026-08-29). Card → pm:queue.

---

Generated by Claude Code