[objectstack-fleet]

Fixes #25767
Supersedes #26138

Clause-②: no (narrowing)
Metadata-protocol: narrowed
Skills-docs: follow-up #26303
Release: 18.3

维护者速读(草稿)

• 改了什么: 按 spec: retire tombstones, conversions and migration entries older than one protocol major; a strict unknown-key refusal with an upgrade-guide link takes over (maintainer ruling D2 = A) #25767 的裁决(6434038284):≤16.x 的退役套件(50 个 D2 conversion 和它们的 fixture、strict 容器里的 55 个墓碑、authorable-surface 上的 61 行 [RETIRED])从代码里删掉。被删的键由 strict schema 当未知键拒绝,错误里多一句话,指向协议升级指南。迁移链不动:支持底线仍是 16,第 17 步和它的 159 个条目文件都在,os migrate meta --from 16 照常从 16 重放到 18。
• 为什么改: 每个 major 都在累积墓碑和 conversion,每张 spec 卡都要为它们买单。升级者需要的处方放在升级指南和当时的 release notes 里就够了。链的底线是发版策略,要另行裁决,不随套件老化一起动(feat(spec,cli)!: retire the ≤16.x retirement kits; the strict unknown-key refusal names the protocol upgrade guide #26138 原先连链一起砍,已按裁决撤回)。
• 风险与回滚: ① 仍写着 ≤16.x 键的源码不再拿到逐键处方,只拿到未知键拒绝和链接。② sys_metadata 里仍是这种形状的行,读取时不再被改写,要在 18.2 或更早版本上用 os migrate meta --stored --apply 规范化一次(裁决把这条一次性提示放进 18.2 的 release notes)。③ 第 17 步的 graduated 列表少了 45 个已删 conversion 的 id,所以 --from 16 不再替你改这 45 类写法;步本身和它的语义条目都还在。回滚就是 revert 本 PR,不写数据,没有不可逆动作。
• 先看: ① packages/spec/src/migrations/registry.ts.template 第 17 步的 conversionIds(删掉 45 个 id 的那一段);② docs/adr/0087-metadata-protocol-upgrade-contract.md 末尾的 2026-10-26 addendum 和 D2 窗口那条修订;③ packages/spec/scripts/build-schemas.ts 的 proof 1(一个 major 的时钟和 release-tag 定日期)。
• 两分钟验证: cd packages/spec && pnpm exec vitest run src/migrations/ src/shared/retired-key-migrate-sentence.test.ts,预期全部通过:链的完整性钉子(每个 id 都是真 conversion)、从底线 16 的重放,以及新旧两种处方句式。
• 需要你做的: 以 GOVERNED_APPROVERS 身份批准本 PR(Tier H:docs/adr/** 原地修订)。

What changes

Maintainer ruling 6434038284 on #25767 (2026-10-27, 「同意」) scopes the D2 = A ruling of 2026-10-25. It keeps kit aging and withdraws the chain cut, and the release target is 18.3. This PR is #26138's branch reworked to that scope. #26138 is closed as superseded (6434144688).

Kept from #26138 (the kit ages out):

• The 50 D2 conversions with retiredAfter at or below 16.x (5 at 15.0.0, 45 at 16.1.0) are removed, with their fixtures.
• The 55 strict-container tombstones already [RETIRED] at @objectstack/spec@17.0.0 are removed, along with 61 [RETIRED] lines of the authorable surface and their liveness rows.
• The 58 strip-container tombstones of the same age stay, because deleting one would turn a refusal into a silent strip. The four inline-credential refusals are not retirements and are untouched.
• The strict unknown-key refusal carries RETIRED_KEY_UPGRADE_GUIDE_SENTENCE (shared/protocol-upgrade-guide.ts) after its fix channels, keeping the strictObject 的 history 句夹在「哪个键错了」与「该写什么」之间,在单行 error 显示位上把修法推到 222 字符之后(#5762 实测) #5955 order.
• The deletion gate ages a tombstone one major. Proof 1 dates a deleted line by an exact RETIRED_KEYS_BY_MAJOR row or by the previous major's release tag.
• The console's own refusals in ui/packages/types cover the keys it used to inherit as tombstones: the five ListView keys and the three flat public-block spellings.
• The lint, metadata-core, runtime, metadata-protocol, objectql, services and ui/** test follow-ups are kept.

Restored to origin/main (the step stays):

• packages/spec/src/migrations/entries/**, including the 159 major-17 files, the README and the conversionIds links of the three protocol-18 entries.
• registry.ts.template with MIGRATION_SUPPORT_FLOOR = 16, chain.ts, migrations.test.ts and semantic-relevance.test.ts.
• packages/cli/src/commands/migrate/meta.ts and its twelve tests, and docs/adr/0142.
• The floor prose in docs/protocol-upgrade-guide.md and content/docs/upgrading.mdx. The guide pointer gains one sentence saying where an aged-out kit's prescription lives.
• ADR-0087's "permanent chain", "forever artifact" and 2026-08-01 amendments are withdrawn, and the D3 text stands.

Reworded:

• ADR-0087: the status line, the D2 window bullet, the 2026-07-31 stored-row note and the addendum now say the kit ages out and the step stays.
• The changeset: minor for spec, patch for lint and metadata-core. The cli is no longer touched in published code.
• Retirement house sentences, which follow the conversion rather than the floor:
  • An edit a surviving step-17 conversion still applies (the twelve toMajor: 17, retiredAfter: 17.7.0 ones: the flow-node aliases and lifts, the mongo alias, the retry default, the sharing-rule full → edit rewrite, the header subtitle alias and the inline action params) goes back to main's "Run os migrate meta --from 16 …" sentence.
  • An edit whose 16.x conversion left says "This edit left os migrate meta with the protocol-16.x conversions: apply it by hand, or run the tool on a release that still carries the conversion", followed by the guide link.
  • No sentence names a floor of 17.

Acceptance notes

1. Step 17's graduated list named the aged-out conversions. The ruling's split reading checked that no semantic entry links a removed conversion, and that holds. Step 17's own conversionIds list in registry.ts.template, however, named the 45 conversions retired at 16.1.0, so restoring it verbatim failed the chain's integrity pin ("every step references only real conversion ids").
   • The 45 ids leave that list, with a comment naming the ruling. The step, its 77 semantic entries and its retired-key and retired-def rows are unchanged.
   • --from 16 still replays 16 → 18, but it no longer applies those 45 edits. Each one's tombstone or guide entry names the hand edit.
   • The ADR addendum and the changeset say so.
2. Pins re-pointed by hand, because each replayed an aged-out conversion:
   • The floor replay in migrations.test.ts now uses sharing-rule-access-level-full-to-edit and flow-node-crud-object-alias.
   • The cli cross-hop pairing pin in meta.report-order.test.ts now uses the sharing-rule conversion.
   • The --from 16 shape in migrate-meta-protocol-range.test.ts is now a sharing rule.
   • migrate-meta-write.test.ts takes feat(spec,cli)!: retire the ≤16.x retirement kits; the strict unknown-key refusal names the protocol upgrade guide #26138's protocol-18 fixtures at --from 16. Its aged-out compareTo control now says the conversion left, not the floor.
   • The three repo-project pins on RETIRED_KEYS_BY_MAJOR[17] go back to main.
3. The optional simplification is not taken. The ruling allowed dropping the release-tag dating source if proof 1 could date every deleted line by its registry row. It cannot: 46 of the 61 deleted [RETIRED] lines predate RETIRED_KEYS_BY_MAJOR, and no row names them. The tag source and its four SKILL.md lines stay, at net zero lines (337 → 337).
4. Stored rows. A sys_metadata row carrying a spelling whose conversion aged out is no longer rewritten at read. Canonicalize it with --stored --apply on 18.2 or earlier. app-hidden-to-unpublished is among the removed conversions, so the artifact door's refusal of it by id is removed as moot. The 18.2 release-notes notice is the ruling's part 3 and is not in this diff: content/docs/releases/ is release-owned.
5. Residuals unchanged by this rework, as listed on the card:
   • The guidance tables are not kit rows, and whether they age with the kits is still open.
   • pendingDataMigrations is moot again while the chain still crosses into 17.
   • skills(objectstack-upgrade): the migration floor is 17 and 16.x-and-older rows are no longer rewritten at read, once the 16.x retirement kits leave #26303 (skills) is handled by its own seat under the ruling.

Verification

All readings are at head 393c768582, which carries a merge of origin/main at add806c5a6.

Suites (local, under the verify lock)

• pnpm --filter @objectstack/spec test: 728 files and 21889 tests passed, 1 todo.
• pnpm --filter @objectstack/spec test:repo: 80 files and 1295 tests passed.
• pnpm --filter @objectstack/cli exec vitest run --project unit: 337 files and 4904 tests passed. The integration tier is declared to CI.
• metadata-protocol: 21569 passed and 19 skipped. runtime: 7423 passed and 19 skipped. objectql: 8789 passed. service-automation: 2708 passed. service-datasource: 975 passed. These ran at the tree of 76dad1ceeb. lint (7389 passed) and metadata-core (823 passed) ran at 002710c1dc. The later commits touch none of their files.
• pnpm --filter @object-ui/types test: 393 files and 10696 tests passed.
• cd ui && pnpm exec vitest run scripts/: 177 files passed and 1 failed. The failure is check-i18n-dead-keys "keeps every key it makes reachable OUT of both tiers", a 15 s timeout on the real-tree scan under full-suite load, the same on three runs. It passes when its file runs alone (check-i18n-dead-keys plus check-vi-mock-inherit: 217 tests passed), and the diff touches no i18n source.
• Type-check of types, react, plugin-list, plugin-view, components, core, app-shell and console: all exit 0.
• pnpm --filter @objectstack/dogfood test, the full battery at the final head: 300 files passed, 3 skipped. 2494 tests passed, 18 skipped. Lock verdict command-exit 0.

Generated artifacts and gates

• pnpm --filter @objectstack/spec check:generated: all 14 artifacts current. node scripts/check-regen-pending.mjs: exit 0.
• node scripts/gen-sdui-manifest-node.mjs then check-sdui-manifest: sdui.manifest.json byte-identical (sha256 1f6e1ccb925a…, 106 components).
• ui check:spec-symbols: exit 0.
• The derived gate sweep (dispatch-gates.mjs --commands): 150 derived, 150 run, 0 NOT-MEASURED. 149 exit 0, and 1 exit 1, which is pnpm --dir ui exec vitest run scripts/ with the timeout named above. Reconciled with --ran.
• Merges of origin/main went through scripts/pm/os-regen-merge.sh. Each time the driver took main's side of a generated surface, the branch's own aged-out [RETIRED] lines were deleted again before gen:schema: three in data.json, one in system.json and five in api.json.

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

This PR changes 4 package(s): @objectstack/driver-sql, @objectstack/lint, @objectstack/metadata-core, @objectstack/spec, touching 348 documentable anchor(s). ⚠️ 36 changed file(s) yielded no anchor (packages/lint/src/validate-object-field-refs.ts, packages/lint/src/validate-security-posture.ts, packages/spec/api-surface/shared.json, …), so the pages documenting them are NOT COVERED by this run — this is not a clean bill of health for those files.

106 hand-written doc(s) name something this change touched — list omitted above 15 rows. Re-derive on the tree named below: node scripts/docs-audit/affected-docs.mjs --json 7c34e5dc211fd54f86db04a650916950301ddfd8.

⛔ 18 release-owned page(s) also affected — read-only, see AGENTS.md Documentation Guardrails.

What this run could not see

• 36 changed file(s) yielded no anchor (packages/lint/src/validate-object-field-refs.ts, packages/lint/src/validate-security-posture.ts, packages/spec/api-surface/shared.json, …) — pages documenting those are invisible to this run
• 2 anchor(s) matched too much of the corpus to be a work list: created_at (literal, 39 pages), requiredPermissions (literal, 32 pages)
• 18 name(s) were too generic to anchor anything (single lowercase words)
• the SDK route bridge reached 55 of 207 client-bound route-ledger rows — the other 152 have no registrar path: tail to select them, so pages documenting THEIR client methods cannot appear above, on this or any run. Of those 152: 0 are remediable by widening that discovery convention (an in-repo file declares the path; the convention did not scan it); 55 are structural — on a ledger where NOT ONE row is declared in-repo, so no discovery change reaches them at any price; 97 are undecided (no in-repo declaration, on a ledger that has other in-repo registrars — absence and an unreadable spelling are not distinguishable here). The rows themselves: node scripts/docs-audit/affected-docs.mjs --bridge-coverage
• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.
• a key NAME is not a key, so the hand re-read the line above prescribes can land on the wrong schema. The same spelling is authorable on one governed type and a [REMOVED] tombstone on another for each of active, aria, joins, objects, template, tools and version (censused on [finding] tools is a key on BOTH AgentSchema (tombstoned, dead) and SkillSchema (live, cloud-attested), so a name-based search attributes skill examples to the agent key — it produced a false stop-the-line alarm on PR #19059 #19093 over the liveness ledger's governed types, top-level keys); nothing in a search result distinguishes the two, so a grep hit on a LIVE example reads as evidence about the DEAD key. Measured on fix(spec): the agent.tools liveness row says dead — it claimed live on a key the schema tombstoned #19059: content/docs/ai/agents.mdx was reported as contradicting the agent.tools tombstone over its tools: example at :161, which is inside the defineSkill({ block opened at :155 — the page was already correct. Settle ownership by PARSING the value against both schemas, never by the name: that literal PASSES SkillSchema, and as an AgentSchema it FAILS at tools with the tombstone prescription. ⛔ These names are not the whole class — a key retired through a .strict() guidance map leaves no tombstone in the walked shape and none of them here (tool.category, live as AIToolDefinition.category).

Coarse fallback — 149 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json 7c34e5dc211fd54f86db04a650916950301ddfd8 → packageMentionDocs.

Which tree this was computed on

This run read content/docs from 839b466eb9f1dc2402b888d6ba21c4593c34f619 — the merge of head a0e848fb6e69a29cbe48b2820b2c7a9ffd3d3bb4 into base 7c34e5dc211fd54f86db04a650916950301ddfd8, which is what actions/checkout gives a pull_request run. Not the PR head.

A worktree cut from an older main holds a different content/docs, so re-deriving there can legitimately return a different list — that is a different tree, not a wrong row. To answer on the same tree:

# while this PR is open — GitHub drops the merge commit once it closes
git fetch origin 839b466eb9f1dc2402b888d6ba21c4593c34f619 && git checkout 839b466eb9f1dc2402b888d6ba21c4593c34f619
# afterwards, rebuild it from the two parents, which stay fetchable
git fetch origin 7c34e5dc211fd54f86db04a650916950301ddfd8 a0e848fb6e69a29cbe48b2820b2c7a9ffd3d3bb4 && git checkout -B drift-repro 7c34e5dc211fd54f86db04a650916950301ddfd8 && git merge --no-ff a0e848fb6e69a29cbe48b2820b2c7a9ffd3d3bb4

node scripts/docs-audit/affected-docs.mjs --json 7c34e5dc211fd54f86db04a650916950301ddfd8

⚠️ That checkout carried uncommitted changes, so the commit above does not fully identify what was read.

Advisory only, and a precision-first one (#9192): a page is listed because it names a
symbol, wire route or SDK method this diff touched — not because it mentions a changed
package. Each row says which anchor put it there, so a wrong row is reportable rather than
merely annoying. To re-verify, run the docs-accuracy-audit workflow scoped to these files:
node scripts/docs-audit/affected-docs.mjs 7c34e5dc211fd54f86db04a650916950301ddfd8 → pass the list as
args.docs, on the commit named under Which tree this was computed on.

[github-actions]

⛔ merge queue 构建失败 — 先分诊，再决定要不要重排

队列构建 40493270857 红了。队列跑的是全量套件（PR 侧 CI 只跑 affected 子集），
所以失败的测试可能在本 PR 没碰过的包里 —— 那不是重排能修的。每次盲目重排都会让排在后面的所有 PR 重建一轮。

分类：failure —— 按下面的日志分诊。

失败的 job（日志抽取，best effort）：

• Test UI (1/16) — 失败步骤: Run the ui tests (shard 1/16)

  × unmount unsubscribes: a later pulse reads nothing 23ms
  FAIL   dom  packages/plugin-chatbot/src/__tests__/ChatbotEnhanced.pendingCountPulse-26471.test.tsx > objectstack#26471 — the draft card re-reads its pending count on the metadata pulse > unmount unsub
    ↳ 失败原因: Error: expect(element).toHaveTextContent()
  

↳ 失败原因 是判读的关键：超时（Test timed out in … / Hook timed out in …）多半是负载/时序，不是本 PR 的回归；
断言（AssertionError: …）才指向真实的行为改变。两者的 FAIL 行长得一模一样，只有这一行能区分。

⚠️ 断言这一侧有一类例外，判据是断言在测什么，不是它是不是 AssertionError。 断言的对象是产品行为（一个值、一个形状、一次拒收）⇒ 照上面读：真实的行为改变，去查，⛔ 不要重排掉；
断言的对象是这次实验自身的有效性前提（跑完的耗时、负载下的先后、任何只在时间预算内才成立的条件）⇒ 它跟超时是同一类，同样对负载敏感，重排一次是合法的判别手段。
识别是机械的：断言的消息或它比较的值本身点名了一段时长、一个时间戳、一个耗时计数。实测过的一对 —— AssertionError: SecurityPlugin.init() ran: expected false to be true 测的是产品行为（真回归）；
AssertionError: this run took over a second, so second-precision stamps could have differed too: expected 1006 to be less than 1000 测的是实验前提：它守护的那条不变式当时是绿的，同一个 head 原样重排一次即成功。
穿着 AssertionError 外衣的时间测量，仍然是时间测量。（⛔ 这只改「怎么读一次红」，不改「哪些测试可以重排」——后者由别处管。）

跨 PR 相同签名（24h，按失败测试文件聚合）：

• packages/plugin-chatbot/src/__tests__/ChatbotEnhanced.pendingCountPulse-26471.test.tsx — 24h 内已弹出 2 个 PR（chore(ui)!: stop publishing @object-ui/app-shell; delete the two host-console examples and the extraction guide (#25922) #26483、feat(spec)!: retirement kits whose conversion retired at or below 16.x age out; the strict unknown-key refusal names the protocol upgrade guide #26599），但它们同属 1 条投机栈 ⇒ 1 次独立命中。汇总 issue：Queue-flake anchor: packages/plugin-chatbot/src/__tests__/ChatbotEnhanced.pendingCountPulse-26471.test.tsx #26647（新建）
• ⚠️ 24h 评论账本没读完（超过 5 页仍未读到窗口尽头），所以上面的「不同 PR 数」是下界，不是全量。

历史信号：

• 本 PR 过去 24h 无队列失败记录（首次）。
• 过去 24h 队列共有 5 个失败构建（不含本次）。

分诊清单：

1. 失败测试在本 PR 改动的包里 → 真回归，修 PR。
2. 失败测试与本 PR 无关 → 看上面的「跨 PR 相同签名」；已有汇总 issue ⇒ flaky/环境问题实锤，去那张 issue 上谈，修好前重排只会再烧一轮全队列。
3. 两者都不是 → 可能与同组 PR 语义冲突；等前面的 PR 落地或失败出队后再重排一次即可，不要连续重排。

---

Generated by Claude Code · merge-queue-triage workflow (#4859)