[objectstack-fleet]

Requested by Leehom · project thread

Part of #23181 (slice 15). The card is a series and stays open.

Clause-②: no
Metadata-protocol: none

Test-only, comment-only diff: skip-changeset. Ten lines under packages/cli/test/, which is not in @objectstack/cli's files[] (dist, README.md, CHANGELOG.md); no files[] changes in any package.

维护者速读

• 改了什么:给 cli 里 10 个「真实启动 os serve / os dev / os start」的测试文件各加一行第 2 行的 // test-budget: 声明,只是注释。
• 为什么改:test(cli): write oclif.manifest.json once per integration run, so a CLI spawn reads its command table #26256 把每次 CLI 启动从约 8 s 降到约 2 s 之后,这 10 个文件仍超 10 s;它们的耗时就是每个独立夹具各启动一次服务,没有可共享或重复的启动,按裁决 6181973363 B 声明实测预算。
• 风险与回滚:零运行时风险,不改断言、超时与分层;回滚即删这 10 行。
• 先看:packages/cli/test/serve-config-boot-residual.integration.test.ts、packages/cli/test/artifact-flag-precedence.integration.test.ts 第 2 行(预算最大的两行)。
• 两分钟验证:node -e 用 scripts/measure-test-shard-timings.mjs 的 readTestBudget 解析这 10 个文件,预期 10 行都返回 spawns the real producer once per single-cause fixture、行号 2。
• 需要你做的:无,席位落地。

What changes

One header line at line 2 of each file below, the ruled grammar // test-budget: N — spawns the real producer once per single-cause fixture: REASON (ruling record 6181973363, letter B; grammar readTestBudget in scripts/measure-test-shard-timings.mjs). N = slowest CI reading x 1.25, rounded up to the next 10 s. Nothing else changes: no assertion, no timeout, no tier, no helper.

file	spawns (counted from the code)	CI run 40392443783 (s)	CI run 40424746941 (s)	local, 3 runs (s)	budget
test/artifact-flag-precedence.integration.test.ts	10 boots: 3 os dev, 7 os start (each a supervisor plus its serve child)	166.18	206.80	152.9 / 153.2 / 138.3	260
test/serve-config-boot-residual.integration.test.ts	17: 8 served os serve boots, 2 refused boots, 5 os build, 2 os validate	191.07	190.07	140.5 / 142.9 / 128.8	240
test/serve-package-declared-capabilities.test.ts	5 os serve boots, one per case and fixture	68.81	43.26	49.7 / 48.1 / 47.0	90
test/serve-config-boot-flat-docs.integration.test.ts	3 os serve --dev boots, one per fixture	47.95	30.34	35.6 / 33.9 / 33.2	60
test/serve-view-container-read.integration.test.ts	3: config boot, os build, artifact boot	35.48	36.70	29.3 / 26.3 / 25.6	50
test/serve-host-config-metadata-door.integration.test.ts	2 os serve boots of one host config (source, compiled twin)	26.55	34.85	29.1 / 25.0 / 25.1	50
test/serve-no-auth-secret-refusal.integration.test.ts	10 os serve boots, one per case (5 refused before the kernel, 5 reach ready)	25.87	31.59	24.7 / 22.8 / 24.2	40
test/serve-migrate-exit-schema-parity.integration.test.ts	2 migrate-and-exit runs (server, --no-server)	30.02	19.15	22.9 / 20.5 / 21.3	40
test/dev-no-watch.pin.test.ts	3 os dev spawns, each a distinct argv and directory	24.16	17.74	23.9 / 22.2 / 23.5	40
test/dev-connector-env-parity.integration.test.ts	3 boots, one per port leg	14.45 (old 3-case shape, not used)	16.35	17.7 / 17.9 / 17.6	30

Why

The reading. No shard-timings refresh has run on main since slice 14 (PR #26256, 9c922b55c1); the newest refresh is run 40275747937, which is older. So the CI numbers above are the vitest module lines (they include beforeAll) of the only two scheduled full-battery ci.yml runs on main after it: 40392443783 (head 69aa630b9f) and 40424746941 (head e843b5b320). Both trees carry packages/cli/test/helpers/oclif-manifest.global-setup.ts, so every spawn there already has the manifest's 2 s floor. The full re-measured offender table (cli from both runs, plus a local run of the other 64 domain:server packages) is in the slice report on #23181.

Why these ten are one family. Each file's wall is real served boots of the CLI, one per distinct fixture or flag, already sequential in beforeAll hooks or one per case. Read per file: no boot is repeated on the same fixture and flag, and no two cases could read one boot. So the two recipes that cut time do not apply. Hoisting boots into a shared beforeAll gains nothing because the cases are distinct fixtures. Sharing a fixture would merge distinct fixtures. What is left is the spawn floor. The local per-case lines show it: in serve-package-declared-capabilities each case is one boot of 7.3 to 12.4 s, and in serve-no-auth-secret-refusal each case is one boot of 1.6 to 3.6 s.

Why not the other boot files. dev-boot-output-order.integration.test.ts boots one fixture several times on purpose (one order "on every boot"). package-restart-acceptance.integration.test.ts and dev-standalone-self-heal.integration.test.ts boot one database twice because the restart is what they test. The class text "once per single-cause fixture" would be false for those three, so they are left for their own reading. serve-runtime-state-project-key.test.ts spawns a writer child, not the CLI.

Tests

Comment-only diff, so behaviour and wall are the tree's. A/B of a comment line measures nothing, so the after reading is the same three runs on this head's tree.

Local runs of the family (N=3, one lock hold)

cd packages/cli && pnpm exec vitest run --project integration --maxWorkers=2 --reporter=default plus the ten files, three times in one os-verify-lock hold, on a shared 4-vCPU container, manifest setup on (queue tier). Each run: Test Files 10 passed (10), Tests 71 passed (71), exit 0. Run walls 297.2 / 285.2 / 270.8 s. Sum of file lines 526.4 / 512.8 / 484.7 s.

Header parse

All ten lines parse through readTestBudget (exit 0): each returns class spawns the real producer once per single-cause fixture at line 2, seconds 260 / 240 / 90 / 60 / 50 / 50 / 40 / 40 / 40 / 30.

Typecheck and lint

pnpm --filter @objectstack/cli typecheck: VERDICT command-exit 0 (check:test-typecheck: OK). pnpm exec eslint --no-inline-config on the ten files: exit 0. node scripts/check-nul-bytes.mjs: exit 0.

Acceptance notes

• Pin: the next shard-timings refresh lists none of these ten files in its over-budget table. Each appears under the budgets they declare, and comes back only if a run reads it over its own budget.
• Slice 11 (PR test(cli): declare measured budgets on the generate-refuses and build spawn files (#23181 slice 11) #26174) headers whose files read under 10 s in both runs above, for the next refresh to drop: test/generate-refuses-retired-generator.test.ts (9.95 / 9.67 s) and test/generate-refuses-name-outside-charset.test.ts (8.77 / 6.49 s). Under 10 s in one run only: -namespace-prefix (13.11 / 8.61) and -unparseable-name (12.37 / 7.82). Still over in both: -unimportable-alias and the five build-* files. Other slices' headers under 10 s in both runs: serve-plugin-auth-load-failure-refusal (9.86 / 7.15) and serve-datasource-mapping-before-sync (7.64 / 5.08). ⛔ Not lowered by hand, per the order: a refresh re-sizes them.
• The budgets rest on two CI readings. A refresh over its 20-run window may read one of them higher, and that is the refresh's job.

Verification

Local gates at 0863e1aa3: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derived 54 commands. Each one ran with its exit code captured before any pipe, and all 54 exit 0. One of them, check-plugin-teardown-shape.mjs --self-test, first exited 3 on the shallow clone; it exited 0 after git fetch origin 621a487607. --ran reconciliation: 54 derived famil(ies) accounted for — 54 run, 0 NOT-MEASURED, exit 0. The repo-wide pnpm lint is left to CI.

Human Minimum

• 1. Schemas start as Zod; packages/spec holds no runtime logic. (not touched)
• 2. Every name I added uses the casing for its kind. (no names added)
• 3. No formatter ran; pnpm lint passes. (eslint on the touched files: exit 0)
• 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. (publishes nothing: skip-changeset)
• 6. packages/spec touched: its generated artifacts are regenerated and committed. (not touched)
• 7. No hand edits to content/docs/references/, packages/*/CHANGELOG.md or content/docs/releases/.
• 8. Title, body and comments are in English. (the maintainer brief is in Chinese, per AGENTS.md)
• The Before Submitting list in CONTRIBUTING.md is done.

🤖 Generated with Claude Code

https://claude.ai/code/session_01FEkQ5nzM2GqFAcHGd5cWx4

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

Nothing in this diff resolved to a documentable surface (no symbol, route or SDK anchor derived from 0 changed package(s)), so this run has no opinion about the docs.

What this run could not see

• 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 — 0 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 75afeffdf564ae0bd4c0baa18791f765d10a91a8 → packageMentionDocs.

[objectstack-fleet]

席位核对 · ACCEPT · domain:server seat 1 · os-project-manager · session_01R1q2xdtYgU6FKrZ7LgjPCC · 2026-10-27T04:12Z

Head 0863e1aa32, report 6433382862 (on #23181), claim 6432333322. No contract review is owed: there is no packages/spec/** diff, no governed path and no narrowing.

item	reading
Shape	Draft, base main, Part of #23181 (slice 15), skip-changeset (test-only, no files[] change). Not governed (0 of 10 paths), 10 changed lines.
Fix	Measure first, as ordered. No shard-timings refresh exists after #26256, so the offender table comes from the two scheduled CI runs after it (40392443783, 40424746941); the evidence copy is at pm-orders/server/evidence/23181-s15/offenders-after-26256.md. The largest remaining family is 10 packages/cli files whose wall is real os serve / dev / start boots (with their build / validate spawns), one per distinct fixture, with nothing shareable. Each gets the ruled // test-budget: header at line 2 (ruling 6181973363 letter B).
Seat check	All 10 headers sit at line 2, and each N is its slowest CI reading × 1.25 rounded up to 10 s. The seat recomputed all ten: 206.80→260, 191.07→240, 68.81→90, 47.95→60, 36.70→50, 34.85→50, 31.59→40, 30.02→40, 24.16→40, 16.35→30. No assertion, timeout or project placement changes.
Evidence	36 check runs succeeded and 16 were expected skips on this head, 0 failed; mergeable clean. Locally N=3: 10 of 10 files and 71 of 71 tests on each run. Derived gates: 54 of 54 exit 0, with 0 NOT-MEASURED.
Carried	The report names the slice-11 header files that are now under 10 s after #26256. The next shard-timings refresh can drop their headers; this PR lowers none, as ordered.
Landing	「按照 pm-dispatch 规则复核过的都授权」 (Leehom, 2026-10-25T11:35Z): ready and auto-merge through the relay, once every check on this head is green or an expected skip.

---

Generated by Claude Code