[objectstack-fleet]

Requested by Leehom · project thread

Part of #23181 (slice 12, claim 6417640032). The card stays open: this PR cuts real work in five of the eight packages/cli/src/commands/migrate/*.integration.test.ts rows, and closes none of the 10 s rows outright. The routes that could close them are in "Not taken" below.

Clause-②: no
Metadata-protocol: none

skip-changeset: a test-only diff. The five files are *.integration.test.ts under packages/cli/src/, and @objectstack/cli publishes files: [dist, README.md, CHANGELOG.md]; the build program does not compile test files.

The reading used

The newest refresh with a per-file list is PR #25960 (14 hourly full-battery runs, measure-test-shard-timings --diff); #26089 landed from the same list. This family's rows in it, CI wall per file:

file	CI s (run)	spawns of the CLI per file
data-commands.absent-database	394.05 (40200279306)	43 CLI + 2 seed boots
object-scope	180.37 (40204927516)	18 CLI + 1 seed + 17 reader/writer children
plan.boot-parity	118.02 (40204927516)	6, through tsx
plan.telemetry-sibling	113.36 (40200279306)	5, through tsx
unmapped-columns	98.92 (40204927516)	10 CLI + 1 seed + 2 reader children
plan.declared-default-datasource	95.65 (declared 80)	4, through tsx
plan.nullable-tenant-columns	62.72 (40204927516)	4 (1 serve boot, 3 plans), through tsx
resume.recorded-by	43.98 (40200279306)	4 crash children + 2 serve boots, resume in-process

The family's ninth row, plan.deferred-reads (18.22 s), is left for a later slice: the claim caps a slice at about 8 files.

Mechanism: the order's hypothesis does not hold here

The order's usual shape, a per-case boot, is not present. Every file already builds its fixture once, in a beforeAll it owns or in a case that owns its fixture alone. Each file's cost is the number of CLI spawns times the per-spawn floor, run one after another. On this box (4 cores, idle, inside the verify lock):

• node bin/run-dev.js --version: 4.93 to 4.99 s; migrate resume --json on an absent database: 5.25 to 6.06 s.
• tsx bin/run-dev.js --version: 8.28 to 8.44 s; the same migrate resume: 9.35 to 10.09 s.
• One node -e child that opens the database with driver-sql and reads it: 0.45 to 0.54 s.

Most of the 5 s floor is oclif importing every command from src/, because no oclif.manifest.json ships for the source entry. That is product code and outside this claim.

What changed

1. Four files spawn the source entry with this process's node, not the tsx binary (plan.boot-parity, plan.telemetry-sibling, plan.declared-default-datasource, plan.nullable-tenant-columns). The entry (bin/run-dev.js), its src/ and the NODE_ENV=development it pins are unchanged. This is how the family's other spawners already spawn it (data-commands.absent-database, object-scope, unmapped-columns, and resume.recorded-by for serve). The tsx wrapper added a second process and its own transform, about 3.4 s per spawn. Fixture configs still load through the CLI's own bundle-require. vitest-tiers.ts still files all four as integration (they value-import node:child_process and name run-dev).
2. object-scope reads and writes its fixture database in-process through better-sqlite3, as plan.boot-parity and plan.telemetry-sibling already do. It used one node child per read: 16 readState calls and 1 off-shape write. Each query is the same column list and ORDER BY id, and the connection is opened read-write as knex opened it.

Every assertion is unchanged. The diff touches no expect, no case and no timeout.

Measured

Same box, before (a335f1de) and after (54d51cc3), N=3 per file, one file per vitest run --project integration --maxWorkers=1 under os-verify-lock. These are process wall and user CPU, which include the hook (the JSON reporter's per-file time leaves out beforeAll).

file	before wall s (3 runs)	after wall s (3 runs)	median change	user CPU s, medians
plan.boot-parity	63.16 · 67.17 · 63.06	40.54 · 37.84 · 41.89	−36%	87.5 → 56.2
plan.telemetry-sibling	64.47 · 67.21 · 62.03	53.08 · 38.21 · 42.63	−34%	89.6 → 58.4
plan.declared-default-datasource	63.37 · 63.40 · 63.33	48.20 · 47.17 · 51.16	−24%	89.4 → 69.3
plan.nullable-tenant-columns	42.54 · 42.74 · 42.72	28.34 · 29.28 · 27.87	−34%	59.2 → 39.1
object-scope	108.79 · 117.71 · 112.76	106.64 · 100.00 · 105.03	−7%	157.6 → 149.4

CPU falls with wall, so this is work removed, not work packed.

Projection, not proof. Scaling by each file's own CI-to-local ratio (CI reading ÷ local before-median): plan.declared-default-datasource ≈ 73 s, under its declared 80 s budget, so the next refresh should stop listing it. That is the one row this PR is expected to close. The others project to plan.boot-parity ≈ 76 s, plan.telemetry-sibling ≈ 75 s, plan.nullable-tenant-columns ≈ 42 s and object-scope ≈ 168 s, all still over 10 s.

Isolation. Two shuffled runs of the five files, all green:

• --sequence.shuffle --sequence.seed=23181: 5/5 files, 27/27 tests.
• --sequence.seed=7 with an execution-order reporter: file order nullable → boot-parity → object-scope → declared-default → telemetry, and object-scope's 20 cases ran out of definition order; 27/27 passed.

The hook files' cases only read what a hook recorded. plan.boot-parity and plan.telemetry-sibling build a fresh mkdtemp fixture per case.

Not taken: the routes that would close the remaining rows

None of these is inside the claim. They are open questions on the card for the seat.

• A. Spawn the built entry (bin/run.js). Its floor was measured at 2.46 to 2.66 s against 5.45 to 6.07 s (vitest.config.ts); slice 9 reads 2.7 to 3.0 against 10.0 to 10.3 s. It changes the NODE_ENV posture and makes dist/ a prerequisite, so it needs a ruling. Slice 9 raises the same question.
• B. Run a file's independent spawns concurrently. Measured on data-commands.absent-database (a local experiment, not committed): pool 1 → 263.1 s wall, 431.9 s CPU; pool 2 → 152.4 s, 422.5 s; pool 4 → 120.7 s, 433.4 s. Wall falls while CPU stays flat, so this packs the work rather than removing it. vitest.config.ts rejects the same trade for maxWorkers because it spends cores that belong to the shard.
• C. A declared budget (ruling record 6181973363 letter B, class "spawns the real producer once per single-cause fixture"). plan.declared-default-datasource and audit-mail-bodies.archive in this directory already use it. It conflicts with the card's "runs under 10 s".

Gates

All at head 54d51cc3d.

• node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack derived 58 commands over the 5-path change set. Each one's exit code was captured by redirecting, then reading $?, with no pipe. All 58 exit 0. --ran: "Run reconciliation — 58 derived, 58 run, 0 NOT-MEASURED, 0 UNRUN", exit 0.
• Six gates first exited 3 (PREREQUISITE NOT MET), and all six exit 0 on re-run: check-plugin-teardown-shape --self-test (shallow clone; fixed by fetching its positive control 621a4876), and check:dual-build-cjs-loads, check:i18n, check:i18n-coverage, check:i18n-walk-parity, check:type-check-debt (no built dist/; fixed by a full turbo run build, 121/121 tasks).
• eslint --no-inline-config packages/cli: exit 0.
• pnpm --filter @objectstack/cli typecheck (tsc --noEmit + check:test-typecheck): VERDICT command-exit 0.
• The five files, vitest run --project integration (the integration tier is this diff's own layer): 5/5 files and 27/27 tests in every run above (3 ordered runs and 2 shuffled).
• packages/cli/dist after the build contains no test file (dist/commands/migrate has 0 *.test.*).

Acceptance notes

• unmapped-columns has the same reader-child shape (2 children, about 1 s of its about 99 s). It is left as it is: its readers compare BLOB-carrying rows byte for byte, which is not worth the risk for one second.
• data-commands.absent-database, unmapped-columns and resume.recorded-by already spawn through node. Inside this claim their only lever is B, which is not taken.

🤖 Generated with Claude Code

https://claude.ai/code/session_01XYjtBuXuGmMz385DUttmzN

---

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 fbb75998ba0e44fe9c3ee9ff4f000d32ee4f7900 → packageMentionDocs.

[objectstack-fleet]

席位核对 · ACCEPT · domain:server seat 1 · os-project-manager · session_01R1q2xdtYgU6FKrZ7LgjPCC · 2026-10-26T15:55Z

Head 54d51cc3d, report 6419936423 (on #23181, slice 12), claim 6417640032 and its amendments (newest 6420575177). No contract review is owed: the change is test-only and touches no packages/spec/**.

item	reading
Shape	Draft, base main, Part of #23181 (slice 12), skip-changeset, mergeable clean.
Files	5 files, +61/−37, all in packages/cli/src/commands/migrate/*.integration.test.ts, which is slice 12's surface. plan.boot-parity is this family's file, not src/utils/boot-preparation-parity.test.ts (#26147).
Fix	Four files spawn bin/run-dev.js with process.execPath instead of through the tsx binary. That is the same entry, the same src/ and the same pinned NODE_ENV, as the family's other spawners already do. It saves about 3.4 s per spawn. object-scope writes its off-shape value and reads its state in-process through better-sqlite3 (already the sibling's import) instead of 17 node children. Every expect is unchanged. No timeout is raised and nothing is deleted.
Evidence	Local medians (N=3): boot-parity 63 → 41 s, telemetry-sibling 64 → 43 s, declared-default 63 → 48 s, nullable 43 → 28 s, object-scope 113 → 105 s. User CPU falls in step, so the work is removed, not packed. Two shuffle seeds both pass 27/27. 58/58 derived gates exit 0 and --ran reconciles. On this head, 36 check runs succeeded and 16 skipped as expected, with 0 failed.
Seat ruling on the open question	The route for the remaining rows is A, then C. A, the source-entry oclif manifest, is already a seat ruling and is in development as slice 14 (a test-only globalSetup). It is not a separate lane decision. After slice 14 lands, the next refresh reads these rows again. Any row still over 10 s takes a declared budget under ruling 6181973363 B, from that CI reading. Not B (packing removes no CPU) and not D.
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.

---

Generated by Claude Code