[objectstack-fleet]

Requested by Sam · project thread

Closes #31606

Clause-②: yes (widening)
Metadata-protocol: none

维护者速读

• 改了什么：停用 OAuth 客户端后，它已经发出去的访问令牌在下一次调用就被 MCP 入口拒绝（401）；平台管理员吊销别人的 API 密钥时记下「已被管理员暂停」，密钥所有者不能自己恢复，只有平台管理员能恢复。
• 为什么改：实测停用客户端后旧令牌照样能用，直到 1 小时后过期；管理员吊销的密钥，所有者马上就能点「恢复」。管理员的「立刻停掉」两头都不生效。
• 风险与回滚：每次 MCP 调用多一次按 client_id 的索引读取；读取失败按拒绝处理。回滚就是还原这个 PR，不涉及数据迁移。
• 先看：packages/plugins/plugin-auth/src/mcp-token-revocation.ts（mcpConsentAdmits），packages/plugins/plugin-auth/src/api-key-restore-guard.ts（stampSuspension）。
• 两分钟验证：pnpm --filter @objectstack/plugin-auth exec vitest run src/mcp-token-revocation.test.ts src/api-key-restore-guard.test.ts，预期全部通过，其中新增的 card 2 用例共 12 个。
• 需要你做的：无，席位落地。

What changes

OAuth clients: a disabled client's live tokens. The MCP door's per-call grant read (mcpConsentAdmits, mcp-token-revocation.ts) now reads the client row first, through the same better-auth adapter it reads the consent through. A token is refused (the verifier answers null, so the door answers the existing 401) when the client is disabled, when its row is gone, or when the read fails. A server-only skip_consent client's flag is read off that same row, so the consent-less path costs no extra read. The provider already refused a disabled client new tokens (invalid_client at the token endpoint, client_disabled at authorize, inactive at introspection) by the same truthy test on the same column; only this door read none of it. Disabling is a pause, not an end: the consent and the refresh family stay, so enabling the client again serves the same grant. Ending a grant stays the withdraw or /oauth2/revoke. Cost: two indexed reads per MCP call (client by unique client_id, then consent) instead of one.

API keys: an administrator's stop holds against the owner. This corrects the card's premise (agreed in the cloud#3820 PM thread): sys_api_key.revoked is not terminal. The Revoke API Key / Restore API Key row actions already make it reversible, and only organization_membership_ended makes it final. The real gap was that an owner could restore a key a platform administrator had revoked the moment it was revoked. So, per the ruling in that thread (option A, no new column, no new state):

• revoked_reason gains a third value, administrator_suspended (sys-api-key.object.ts, label and description translated in all nine shipped locales).
• The restore guard (api-key-restore-guard.ts, rule 4) stamps it on a user-context revoke by a caller with platform-admin standing who is not the row's user_id, when the row carries no reason yet. A host that stops a key itself (cloud's console, in the system context) writes the same value beside revoked: true; system writes are never stamped or refused.
• Rule 2 extends to it: only a platform administrator restores such a key. The owner's restore is refused 403 PERMISSION_DENIED, naming the reason and both roads. Restore clears the reason, as before, and the same key works with no re-issue.
• An administrator's revoke never replaces a reason the key already carries, so organization_membership_ended stays final.
• A multi-row update shares one payload, so a stamp written for one row lands on every row. A multi-row admin revoke that matches both a row the rule stamps and a row that already carries a different reason is refused whole with 400 VALIDATION_ERROR, in either row order, before the driver writes. A batch of plain keys is stamped on every row. Over-stamping an administrator's own key in such a batch is harmless, because the administrator can restore it.
• The key door's 403 API_KEY_REVOKED text (@objectstack/core api-key.ts) gets an arm that names a platform administrator as the only restorer.

⚠️ Still open, by design, and owned elsewhere: this closes the owner's restore, not the owner's new key. A person who may mint keys can still mint a fresh one, and it acts as them. An AI service account whose credentials only administrators issue closes that gap (cloud#3820 O1/O3, objectstack#31603). This PR does not try to solve it.

Step-1 measurement (before the fix)

Real routes, in-process HTTP, on objectstack 158cc2aa (main when measured): the showcase dogfood stack with MCPServerPlugin, OS_OIDC_PROVIDER_ENABLED=true. Flow: DCR public client, sign-in, authorize (PKCE, resource = the MCP URL), consent, code exchange.

OAuth client disable: the old token kept working

Step	Answer
access-token lifetime	expires_in 3600, exp − iat = 3600 s
POST /mcp initialize, before the disable	200
POST /auth/admin/oauth2/toggle-disabled {disabled: true} as the platform admin	200
POST /mcp initialize, same token	200, served (the defect)
refresh_token grant while disabled	400 invalid_client "client is disabled"
enable again: the same token / its refresh	200 / 200

API key: the owner undid an administrator's revoke

Step	Answer
member mints a key (POST /api/v1/keys), key works	201, then 200
platform admin PATCH /data/sys_api_key/{id} {revoked: true}	200; the key then answers 403 API_KEY_REVOKED
member PATCH /data/sys_api_key/{id} {revoked: false}	200; the key works again (the hole)

Tests

Two commits after this section was written seed a client row in three existing verifier test files (26 cases that read no client row) and update three docs pages; the os-dev-report on #31606 covers both.

New pins, each as a top-level describe in that component's existing harness file:

• plugin-auth mcp-token-revocation.test.ts, on the real AuthManager and the installed provider (4 cases): a token minted before the disable is refused, its refresh is refused by the provider, the grant rows stay, and once enabled again the same token and a refreshed one verify; a disabled skip_consent client is refused; a token whose client row is gone is refused; a store error in the client read refuses. All 4 red on the old verifier (restored from origin/main while the tests stayed): expected { …(3) } to be null.
• plugin-auth api-key-restore-guard.test.ts, on the real engine, SecurityPlugin, D2 guard and key door (8 cases): the stamp, the owner's 403, the admin's restore and the key working again; CONTROL: an admin's own key gets no reason; an owner-revoked key re-revoked by an admin is stamped; an owner's re-revoke and smuggled clear keep it; an admin's revoke keeps organization_membership_ended and admin_password_reset; a host's system suspension holds; a plain multi-row revoke stamps every row; a mixed multi-row revoke is refused whole in both orders. 6 of 8 red on the old guard. The 2 that stay green are the controls that hold today too.
• plugin-auth forced-reset-revoke-reason-vocabulary.test.ts: the option list is exactly the three values, and the description names the new one.
• core api-key-lifecycle.test.ts: the door text for administrator_suspended names only a platform administrator.
• dogfood authorized-client-withdraw.dogfood.test.ts, over HTTP: a member's toggle is refused 403 and moves nothing; the admin's disable makes the same bearer 401 at /mcp and its refresh 400 invalid_client, while the consent stays; enabled again, the same token, a refreshed token and a new person's consent on the same client are all served. Red against the old plugin-auth build (expected 200 to be 401), green after rebuilding.
• dogfood api-key-owner-revoke.dogfood.test.ts, over HTTP: the admin's revoke_api_key PATCH refuses the member's key on its next call and stores administrator_suspended; the member's restore_api_key PATCH is 403 PERMISSION_DENIED; the admin's restore brings the same key back and clears the reason; CONTROL: the member's own revoke and restore work as before.

Both dogfood files had their boot and helpers hoisted from inside the existing describe to module scope, so the new describe shares the one showcase boot. git diff -w shows that move as a few lines.

Acceptance notes

• Declaration lines (seat 3 confirmed the changeset's reading on review: patch for all three packages, precedent objectstack#31435) are at the top of this body.
• Re-enable revives the outstanding grant, the access token and its refresh alike. That is deliberate, because the card's "one-click disable" is recoverable. A client that must lose its grants for good is withdrawn or revoked.
• No new error code. The OAuth refusal is the existing 401. The key refusal stays API_KEY_REVOKED, and the restore refusal PERMISSION_DENIED. The mixed-batch refusal is VALIDATION_ERROR, reachable only from an in-process multi-row user-context write, because sys_api_key has no batch route.
• Who stamps. Only a caller with platform-admin standing (hasPlatformAdminStanding) who is not the row's owner. The shipped permission sets let no one else write another person's key (sys_api_key_self), so in a shipped composition that is every non-owner revoke.

Verification

See the os-dev-report on the card for package suites, typecheck and the dispatch-gate sweep at the final head.

Human Minimum

• 1. Schemas start as Zod; packages/spec holds no runtime logic. (No packages/spec change.)
• 2. Every name I added uses the casing for its kind.
• 3. No formatter ran; pnpm lint passes.
• 4. One issue, one branch off an up-to-date main, one pull request.
• 5. A change that publishes carries a changeset, and both declaration lines are filled in here and in the changeset. (Changeset carries them; the body lines are the seat's, see Acceptance notes.)
• 6. packages/spec touched: n/a.
• 7. No hand edits to content/docs/references/, packages/*/CHANGELOG.md or content/docs/releases/.
• 8. Title, body and comments are in English.
• The Before Submitting list in CONTRIBUTING.md is done.

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

This PR changes 3 package(s): @objectstack/core, @objectstack/platform-objects, @objectstack/plugin-auth, touching 19 documentable anchor(s). ⚠️ 1 changed file(s) yielded no anchor (packages/plugins/plugin-auth/src/auth-manager-mcp-access-token.ts), so the pages documenting them are NOT COVERED by this run — this is not a clean bill of health for those files.

35 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 dced781f8720643deb63d96d2f901452aceadf78.

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

What this run could not see

• 1 changed file(s) yielded no anchor (packages/plugins/plugin-auth/src/auth-manager-mcp-access-token.ts) — pages documenting those are invisible to this run
• the SDK route bridge reached 64 of 227 client-bound route-ledger rows — the other 163 have no registrar path: tail to select them, so pages documenting THEIR client methods cannot appear above, on this or any run. Of those 163: 0 are remediable by widening that discovery convention (an in-repo file declares the path; the convention did not scan it); 59 are structural — on a ledger where NOT ONE row is declared in-repo, so no discovery change reaches them at any price; 104 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 — 51 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 dced781f8720643deb63d96d2f901452aceadf78 → packageMentionDocs.

Which tree this was computed on

This run read content/docs from 80bd44b9796cf10147afb3360855512c6ae3531a — the merge of head 6beaf69cf09607f6f33ce629d6c4dedd886b7085 into base dced781f8720643deb63d96d2f901452aceadf78, 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 80bd44b9796cf10147afb3360855512c6ae3531a && git checkout 80bd44b9796cf10147afb3360855512c6ae3531a
# afterwards, rebuild it from the two parents, which stay fetchable
git fetch origin dced781f8720643deb63d96d2f901452aceadf78 6beaf69cf09607f6f33ce629d6c4dedd886b7085 && git checkout -B drift-repro dced781f8720643deb63d96d2f901452aceadf78 && git merge --no-ff 6beaf69cf09607f6f33ce629d6c4dedd886b7085

node scripts/docs-audit/affected-docs.mjs --json dced781f8720643deb63d96d2f901452aceadf78

⚠️ 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 dced781f8720643deb63d96d2f901452aceadf78 → 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 42507456241 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）：

• Dogfood Regression Gate (1/3) — 失败步骤: Boot example apps and exercise real user flows

  @objectstack/dogfood:test:  FAIL   isolated  test/authz-conformance.test.ts > ADR-0056 D10 — authorization conformance matrix > is a sound conformance ledger (ADR-0060 checkLedger) + the #2567 surface
    ↳ 失败原因: @objectstack/dogfood:test: AssertionError: UNCLASSIFIED surface — add a ledger row (ADR-0060): rest-family:rest-route-ledger.ts:recycle-bin: expected [ Array(1) ] to deeply equal []
  @objectstack/dogfood:test:  FAIL   isolated  test/authz-conformance.test.ts > the ledger-sourced population and its baseline bite > the classified ledger keys really are classified — the baseline is n
    ↳ 失败原因: @objectstack/dogfood:test: AssertionError: expected 42 to be 43 // Object.is equality
  @objectstack/dogfood:test:  FAIL   isolated  test/authz-conformance.test.ts > #2567 — anonymous-deny surface ratchet bites > the real matrix + real discover is sound (baseline lock)
    ↳ 失败原因: @objectstack/dogfood:test: AssertionError: expected [ Array(1) ] to deeply equal []
  @objectstack/dogfood:test:  FAIL   isolated  test/authz-conformance.test.ts > #9083 — a wired transport is admitted only by an `enforced` row > (j) leg 2 — appending the key to the EXPERIMENTAL realti
    ↳ 失败原因: @objectstack/dogfood:test: AssertionError: expected [ Array(1) ] to deeply equal []
  @objectstack/dogfood:test:  FAIL   isolated  test/authz-conformance.test.ts > #9083 — a wired transport is admitted only by an `enforced` row > (k) the `removed` state admits no exit either — all thre
    ↳ 失败原因: @objectstack/dogfood:test: AssertionError: expected [ Array(1) ] to deeply equal []
  @objectstack/dogfood:test:  FAIL   isolated  test/authz-conformance.test.ts > #9083 — a wired transport is admitted only by an `enforced` row > (l) an `enforced` row naming its enforcement site DOES a
    ↳ 失败原因: @objectstack/dogfood:test: AssertionError: expected [ Array(1) ] to deeply equal []
  @objectstack/dogfood:test:  FAIL   isolated  test/authz-conformance.test.ts > #17111 — the matrix docblock's ledger figures are pinned to the ledgers > every figure the docblock states equals what its
    ↳ 失败原因: @objectstack/dogfood:test: AssertionError: expected [ Array(1) ] to deeply equal []
  @objectstack/dogfood:test:  FAIL   isolated  test/authz-conformance.test.ts > #17111 — the matrix docblock's ledger figures are pinned to the ledgers > a stale GROUP count is RED — the #20102 drift (a
    ↳ 失败原因: @objectstack/dogfood:test: AssertionError: the doctoring must land, or this case proves nothing: expected '// Copyright (c) 2026 ObjectStack. Li…' not to be '// Copyright (c) 2026 ObjectStack. Li…' //
  @objectstack/dogfood:test:  FAIL   isolated  test/authz-conformance.test.ts > #17111 — the matrix docblock's ledger figures are pinned to the ledgers > DELETING the figure is RED — the pin cannot be s
    ↳ 失败原因: @objectstack/dogfood:test: AssertionError: expected '// Copyright (c) 2026 ObjectStack. Li…' not to be '// Copyright (c) 2026 ObjectStack. Li…' // Object.is equality
  @objectstack/dogfood:test:  FAIL   isolated  test/authz-conformance.test.ts > #17111 — the matrix docblock's ledger figures are pinned to the ledgers > a SECOND present-tense copy of a figure is RED —
    ↳ 失败原因: @objectstack/dogfood:test: AssertionError: expected '// Copyright (c) 2026 ObjectStack. Li…' not to be '// Copyright (c) 2026 ObjectStack. Li…' // Object.is equality
  @objectstack/dogfood:test:  FAIL   isolated  test/authz-conformance.test.ts > #17111 — the matrix docblock's ledger figures are pinned to the ledgers > the anchor is the PATH, so the dated drift note 
    ↳ 失败原因: @objectstack/dogfood:test: AssertionError: expected [ Array(1) ] to deeply equal []
  

↳ 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）：

• test/authz-conformance.test.ts — 24h 内已弹出 3 个 PR（feat(rest): the recycle bin has REST doors, and a delete answer names the entry to undo (#30237 PR 5) #31595、fix(plugin-auth): disabling an OAuth client refuses its already-issued tokens on the next call, and an administrator's API-key suspension holds against the owner #31622、feat(service-cache)!: a declared multi-node deployment on the redis cluster driver shares one Redis cache, so cache-backed limits hold across replicas #31624），但它们同属 1 条投机栈 ⇒ 1 次独立命中。汇总 issue：Queue-flake anchor: test/authz-conformance.test.ts #31643（已刷新）
• ⚠️ 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 4 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)