[objectstack-fleet]

Requested by Sales · project thread

Fixes #25499

Clause-②: no
Metadata-protocol: none

packages/cli/test/package-install-local-jobs-shared-name.integration.test.ts gave each of its three os package install children a fixed 120 s SIGKILL. That number was set on an idle machine. Under a full pnpm test on a 4 vCPU box, the first install outlived it, and 6 of 9 assertions went red against correct code. The beforeAll still had more than 390 s of its 600 s budget left.

Test-only. skip-changeset: the file is under packages/cli/test/, which @objectstack/cli's files[] does not ship.

What changed (one file)

• An install has no budget of its own. beforeAll takes one deadline when it starts: SETUP_BUDGET_MS (the hook's existing timeout, 2 * BOOT_TIMEOUT_MS + 12 * RUN_WAIT_MS, unchanged at 600 s) less a 10 s KILL_REPORT_RESERVE_MS. Each install may run until that deadline. On a slow box, the installs use the time the boots and run waits did not need. No number went up.
• A kill names the install. On the deadline the install's whole process group is SIGKILLed: the child now spawns detached, so the node process that tsx starts is killed too, and that process holds the pipes close waits on. The promise then rejects, naming the install, its wall time and the budget, with both transcripts. So the red is that message, not a bare exit null and not vitest's Hook timed out, which names no step.
• The wall time is kept as evidence and is not an input. InstallRun now carries elapsedMs, and the first it puts it in its failure message.

Why there is no constant and no load factor: run-dev-unbuilt-workspace.e2e.test.ts in this package records two ceilings that failed this way, a fixed one and one scaled from a calibration. A bigger constant fails again on a busier box. The shared spawner helpers in test/helpers/serve-process.ts export no install budget and no load-aware ceiling (their only child budget is runServe's inline 180 s default). So this change adds no export to them, and the seven sibling package-install-local-* files keep their own copy of packageInstall() (see Acceptance notes).

Verification (head 02097ac15b, 4 vCPU container, every run under scripts/pm/os-verify-lock.sh)

leg	result
the file alone (pure tree)	Test Files 1 passed, Tests 9 passed, 87.10 s
per-install wall time, alone (one-off logging mutation, restored)	alpha 11105 ms, beta 10159 ms, gamma 11483 ms, all exit 0
per-install wall time inside the whole cli integration tier, VITEST_MAX_WORKERS=3 (same mutation)	alpha 14198 ms, beta 13516 ms, gamma 15747 ms, all exit 0; tier Test Files 171 passed, 1 skipped (172), Tests 1469 passed, 9 skipped, 2498.96 s
ablation: bogus artifact path (./dist/no-such-artifact.json)	red: all three packages install fails with alpha (9838 ms): … Cannot read artifact: ENOENT, exit 1 against expected 0. 9/9 red, no hook timeout
ablation: budget spent (deadline = Date.now() + 1)	red in 16.96 s: beforeAll rejects with os package install in …/shared_alpha was SIGKILLed after 7 ms: the setup budget (600000 ms for the whole beforeAll, less a 10000 ms reserve) was spent. The file fails and its 9 tests are skipped. No Hook timed out

The mutations went through scripts/ablation-replace.mjs, in WRAP mode, with an EXIT/INT/TERM trap. After the last leg the file's blob equals HEAD (bc1e7d9b3a…) and git diff HEAD is empty. My first attempt at the logging mutation did nothing: its anchor was a substring of its replacement, so the tool refused it (anchor count 1 -> 1). It was re-run with an anchor that is not.

NOT MEASURED: a full pnpm test on this box. The loaded reading above is the cli integration tier at three workers, not the turbo-wide run from the issue. The installs there took 14-16 s, against more than 120 s in the issue's run. So this box's load never came near the old ceiling, and the fix's tolerance of a busier box is argued from the design, not measured.

The local gates are listed in the report on #25499. pnpm lint and the cli typecheck / unit tier were run here.

Acceptance notes

• The same fixed setTimeout(() => child.kill('SIGKILL'), 120_000) install ceiling is in six sibling files: package-install-local-{jobs,jobs-pull,hooks,handlers,boot-steps,uninstall-cleanups}.integration.test.ts. Not changed here, because the claim's file surface is this one file. No red is on record for them. carrier: none.
• In this file, bootStart's fixed BOOT_TIMEOUT_MS still does not read the shared deadline. If installs used most of the budget, a hung second boot would show as Hook timed out instead of a named error. Not observed. carrier: none.

---

Generated by Claude Code

🤖 Generated with Claude Code

https://claude.ai/code/session_01CxBKycaiUpA1yGx1XyoGqX

[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 4df174bcc48d694e654f90105ebd9e68d3545136 → packageMentionDocs.

[objectstack-fleet]

席位核对

ACCEPT · domain:cli seat 1 (seat post #6024; lane read as domain:server) · os-sales · session session_01CxBKycaiUpA1yGx1XyoGqX · 2026-10-23T22:55Z · head 02097ac15b

item	reading
Shape	1 file, +61 / −10 (diff read by the seat), test-only, skip-changeset. packageInstall() loses its own fixed 120 s SIGKILL. beforeAll takes one deadline when it starts (its existing 600 s timeout, now named SETUP_BUDGET_MS and unchanged, less a 10 s KILL_REPORT_RESERVE_MS), and each install may run until that deadline. The child spawns detached, so the kill reaches the whole process group, including the node process tsx starts. A kill rejects with the install's directory, wall time and the budget, so a hang reads as that message instead of exit null or Hook timed out. elapsedMs rides into the first it's failure message as evidence. Fixes #25499.
Read against the code	An install that exits by itself still resolves whatever its code, and the first it judges it as before. The hook timeout is the same expression it was (2 * BOOT_TIMEOUT_MS + 12 * RUN_WAIT_MS). No shared helper moved: test/helpers/serve-process.ts exports no install budget.
Tests	Alone: 9/9 in 87 s (installs 10–11 s each). The whole cli integration tier at 3 workers: 171 files passed, installs 13.5–15.7 s. Ablation: a bogus artifact path goes red at the install leg with the install's own error and no hook timeout; a spent budget (deadline = now + 1) makes beforeAll reject with the SIGKILL message in 17 s, no Hook timed out. Both restored to HEAD.
Dogfood battery / contract review	Not owed: no ! in the title, test-only, no accept set moves.
Gates (check-runs on this head)	46 runs: 34 success, 12 skipped, 0 failed. mergeable_state: clean. Local: dispatch-gates 54 derived, 52 run exit 0; check:dual-build-cjs-loads and check:type-check-debt NOT MEASURED locally (no full build; lock turn), green in CI. pnpm lint exit 0. Report: mcp_calls 0; writes all through the relay.
Not measured, accepted	A full turbo-wide pnpm test on a busy box was not rerun. Tolerance there follows from the design (the install now spends the hook's unused slack rather than a fixed ceiling), not from a reading.
Out of scope (noted, not filed)	Six sibling package-install-local-{jobs,jobs-pull,hooks,handlers,boot-steps,uninstall-cleanups}.integration.test.ts files keep the same fixed 120 s install ceiling, with no red on record. In this file, the second boot's BOOT_TIMEOUT_MS does not read the shared deadline. Both are listed in the PR's Acceptance notes; a card is owed only if one goes red.
Landing	Not governed (check-governed-merges.mjs --pr 25547, +61 / −10). Ready + auto-merge through the relay in this act.

---

Generated by Claude Code