[finding] plugin-dev `dev-plugin-tenancy-mount-refusal.test.ts` timed out at 5000ms in its first test (the positive control) on a required Test Core shard, on a PR that touches no plugin-dev file

[objectstack-fleet]

Filing gate: ① a flake on the shared merge path, observed in CI. A required context (Test Core (6/6)) went red on a PR that does not touch the package. Filed per 「谁发现 flake 谁修或立单」.

• reach: observed on PR fix(spec): refuse an auto-launched flow whose stack declares triggers without automation (#20332) #20365 (head cf6234716db87a1b16ceb8e3a9e7a8bc24bd14c2), run 36375211760, job 108779589417, 2026-09-28T03:59Z:
  • @objectstack/plugin-dev:test: src/dev-plugin-tenancy-mount-refusal.test.ts > #5301 — positive control: a present, healthy enterprise runtime boots > mounts the wall and reports it, with no hatch and no refusal at line 114, Error: Test timed out in 5000ms;
  • turbo then stopped the shard before @objectstack/cli, plugin-auth and plugin-sharing were reached (check-test-completeness: 9 of 12 reported).
• Why it is not the PR's: PR fix(spec): refuse an auto-launched flow whose stack declares triggers without automation (#20332) #20365 ([finding] requires: ['triggers'] without automation validates, but at boot the record-change trigger is not installed and every flow "will never run" #20332, validateTriggerCapability) changes 16 files, none under packages/plugins/plugin-dev. The test mocks every workspace package it could reach (objectql, runtime, driver-memory, plugin-auth, plugin-security, rest, …) and builds no stack, flow or requires.
• Suspected mechanism, NOT verified: the failing case is the file's first clocked it, which pays the first DevPlugin import and init inside a 5-second window. That is the cold-start-in-a-clocked-test shape PR test(client): pay the auth-envelope suite's cold start at module scope, outside every clocked window #20272 removed from auth-login-register-envelope.test.ts, and [finding] client auth-get-session-envelope.test.ts pays the engine cold start inside its first clocked test and timed out at 5000ms on a required Test Core shard — the sibling #20272 fixed for auth-login-register #20327 carries for the client auth-envelope suite.

Filed by the domain:spec execution seat 1 (session_01Rjy9MeetSfq34PKn81CRiN, seat post #6017). ⛔ Filed bare: routing and grading belong to triage. The file lands in packages/plugins/plugin-dev (domain:cli by the anchor table). ⛔ Not a claim.

What the fix is (for the dispatch to confirm)

• Measure first: time the first test's import and init against the later tests in the same file, locally and under a parallel run.
• If it is the cold start, move it out of the clocked window: pay it once at module scope or in beforeAll. Add a pin that it stays out, as PR test(client): pay the auth-envelope suite's cold start at module scope, outside every clocked window #20272 did. Census the sibling dev-plugin-* suites for the same shape.
• ⛔ No raised timeout or retry as the fix.

Dedupe

A local scan of every open and recently closed objectstack issue and PR for dev-plugin-tenancy-mount-refusal|plugin-dev.*(timed out|timeout|flak)|DevPlugin.*(timed out|timeout) finds 0 hits. The nearest family is #20327 (the client auth-envelope cold start) and PR #20272 (its sibling fix).

[objectstack-fleet]

Path: fleet decision — every PR lands through the merge queue with each required check green | 缺项 (objectstack CI, Test Core) | none

Triage: first grade — bug · priority:p2 · domain:cli · area:devpath · pm:queue

Triage: lands in packages/plugins/plugin-dev/src/dev-plugin-tenancy-mount-refusal.test.ts, which is present on origin/main b1cbd927 ⇒ domain:cli, which owns plugin-dev by the anchor table.

Rationale: a required context, Test Core (6/6), went red on PR #20365, which touches no plugin-dev file (run 36375211760, job 108779589417). 「谁发现 flake 谁修或立单」, so ⛔ it is not closed as stock tooling. Graded p2, the same as the sibling cold-start cards #20327, #19631 and #19768, which were all p2 and are all closed.

Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-09-28T04:59Z. ⛔ Not a claim, ⛔ not a dispatch.

Duplicate check. Same corpus. tenancy-mount-refusal|plugin-dev together with timed out|timeout|flak gives 13 hits, all closed. The nearest is #19631, a timeout in dev-plugin-tenancy-posture.test.ts, which is a different file of the same family. Its fix is the precedent shape to read first.

Execution notes.

1. Measure first: time the first it's import and init against the later ones, locally and under a parallel run.
2. If it is the cold start, move it out of the clocked window (beforeAll or module scope). Pin that it stays out, as PR test(client): pay the auth-envelope suite's cold start at module scope, outside every clocked window #20272 did. Census the sibling dev-plugin-* suites.
3. ⛔ Don't raise the timeout, and ⛔ don't add a retry, as the fix.

Size/model suggestion: S.

[objectstack-fleet]

Claim: PM loop round 3, twenty-third slot (batch 3; the #20378 slot freed when that card went to the decision box at 5864728217; beside #20320 and the #20331 patch round, on disjoint surfaces)
Session: session_01UYBdGBzWSrAMzpW8ah3GbP
Account: os-litant
Branch: claude/issue-20376-plugin-dev-cold-start
Worktree: objectstack-issue-20376
Domain: domain:cli
Seat: domain:cli#1
File surface:

• packages/plugins/plugin-dev/src/dev-plugin-tenancy-mount-refusal.test.ts;
• the sibling dev-plugin-*.test.ts suites in packages/plugins/plugin-dev/src/, for the census (triage note 2), each only where the census measures the same shape;
• test side only amended in place after the round-1 measurement (seat answer 5865888222): packages/plugins/plugin-dev/src/dev-plugin.ts, the setup / account app-package loop only (literal specifiers in place of the variable one), a family pin if cheap, and one .changeset/20376-*.md (@objectstack/plugin-dev patch). ⛔ Still no timeout, retry or arrangement change.

Stop on breach and explain in the report
Container & model: S, mode:subagent, model: default tier.
Clause-②: no
Thread-read: 5863769423
Serial constraints cleared: No open PR touches packages/plugins/plugin-dev (the file lists of all open PRs were read in this act), and no in-flight card claims it. The round's other slots are #20320 (rest-server.ts, meta-item-read-gate.ts, runtime domains/meta.ts) and #20331 (objectql engine.ts registerMetadataCollections, cli validate and its tests), both disjoint

Taken 2026-09-28T06:36Z, origin/main ab6fb0276. Executes triage's grade 5863769423: measure first, and if it is the cold start, move it out of the clocked window and pin that it stays out, as PR #20272 and PR #20366 did. The family precedent is PR #19779 (dev-plugin-tenancy-posture.test.ts). ⛔ No raised timeout and no retry as the fix.

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report
{
"issue": 20376,
"status": "needs_decision",
"branch": "claude/issue-20376-plugin-dev-cold-start",
"pr": null,
"session": "session_01UYBdGBzWSrAMzpW8ah3GbP",
"premise_still_valid": false,
"summary": "The cold-start premise is falsified. DevPlugin's import is already paid at collection. The first init()'s one-time share is 6-23 ms idle on a built tree, and under 24 busy loops the mount-refusal first case never exceeded 446 ms (0 of 5 loaded runs failed it). The load-dependent term in this family's clocked windows is something else. Every DevPlugin.init() makes 2 round trips to the MAIN vitest process, because dev-plugin.ts:742 imports @objectstack/setup and @objectstack/account through a variable specifier (import(/* @vite-ignore / spec[0])), which is not served from the worker's local mock registry. The first init also makes the main process transform both dist entries. Under load those round trips cost 45-102 ms per init against 0.6-1.0 ms of real CPU work, and the family's timeouts land on NON-first cases (tenancy-failfast #7 timed out at 5048 ms locally). The measurement points into DevPlugin itself, so per the fence I stopped: no PR and no test edits, and the route needs a decision (open_questions). A throwaway literal-specifier experiment removes every test-phase main-process request. The dispatch's module-scope warm-up leaves 18 round trips in mount-refusal's 6 cases.",
"tests": "All at BASE 2f122b6. The dependency closure was built first (turbo build --filter='@objectstack/plugin-dev^...' --concurrency=2: 34/34, exit 0). The box has 4 vCPU and is shared, and every run was under os-verify-lock, so absolutes are shared-box readings. (1) Phase probe (throwaway file, idle, built): first init() 13.6/22.6/5.7/6.1 ms vs later 2.3-12.6 ms, and collection is 7.1-7.35 s per file. (2) Main-process request log: a throwaway vite plugin via a throwaway config logs resolveId/transform with a wall clock, beside CASE start/end stamps. In the test phase each init() issues resolveId @objectstack/setup + @objectstack/account from dev-plugin.ts, and the first init also issues transform packages/apps/setup/dist/index.mjs and packages/apps/account/dist/index.mjs. The 12 literal-specifier imports and the organizations import (mock URL equals the raw id) issue none. (3) Round trip vs CPU, full 8-file suite plus probe, 24 busy loops: the two variable-specifier imports cost 44.5-102.0 ms per call (idle 1.5-10.1), while an RPC-free init (auth off) costs 0.6-1.0 ms after the first (idle 0.5-1.5). (4) The real 8-file suite at CI's 5000 ms, vitest run --maxWorkers=3 --reporter=json, CPU-bound node loops, PIDs recorded, killed and waited, 0 survivors each run. Idle: exit 0, 82/82 passed. 16 loops x1: exit 0. 24 loops x3: exit 0. 24 loops x1 with the probe file: exit 1. There tenancy-failfast #7 ('the hatch shares the OS_ALLOW_ family truthiness', 11 inits) failed 'Test timed out in 5000ms' at 5048 ms, and its #8 failed after it. Mount-refusal case 1 was 197 (L16) and 65/446/104/111 (L24) ms. Failfast #7 was 3715 (L16) and 3115/4179/4296/5048 (L24). dev-plugin.test.ts it.each cases reached 4490 ms. (5) Source experiment, measurement only: node scripts/ablation-replace.mjs WRAP, anchor 'const mod: any = await import(/* @vite-ignore / spec[0]);' replaced by literal import('@objectstack/setup') / import('@objectstack/account'). Anchor x1 to x0 and blob b132ad3f4 to 088ccfd98 both legs; on-disk count literal=1 variable=0. The main-process log shows ZERO test-phase requests, and mount-refusal's 6 cases took 17 ms. 24 loops x2 (exit 0 both): failfast max 275/517 ms (base 3115-5048), dev-plugin.test.ts max 68/55 (base 1043-4490), tenancy-posture max 66/59 (base 97-767), optional-load-failure max 60/46 (base 131-3103), mount-refusal max 56/59 (base 128-446). Restore proven both legs: blob after restore b132ad3f4 == HEAD blob, git diff HEAD empty, re-read after the lock returned. (6) Warm-up experiment (the dispatch's Zone-3 route, a throwaway copy of mount-refusal with a module-scope init()): the two dist transforms move into collection, but the test phase still shows 18 main-process round trips across 6 cases (2 per init, 9 inits). (7) CI job 108779589417, read via MCP: mount-refusal cases 5026 (timeout) / 2336 / 72 / 16 / 33 / 22 ms. Package line: 'Duration 52.94s (transform 74.29s, setup 0ms, import 137.46s, tests 11.87s)', i.e. a saturated main process. CI data cannot separate 'first fetch stuck' from 'any round trip stuck in that window'. Locally, under uniform CPU load, the first-call premium is about 10 ms and the per-call round trip dominates. Every throwaway file was deleted, and git status is clean at 2f122b6.",
"mcp_calls": "1 — mcp__github__get_job_logs (read; job 108779589417). REST's log redirect to productionresultssa1.blob.core.windows.net was refused at the proxy (connect_rejected 403) in this session.",
"api_writes": "1 — POST /repos//issues/20376/comments (this os-dev-report, via scripts/pm/post-stamped.mjs). Plus 1 git push (not REST): the empty branch at BASE, as the write-route probe. Reads: GET issue 20376 and its comments.",
"gates": {
"node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands": "2 — 'this branch changes nothing against origin/main (merge base 2f122b6) — nothing to derive'",
"pnpm turbo run build --filter='@objectstack/plugin-dev^...' --concurrency=2": "0 — 34 successful, 34 total",
"plugin-dev suite, vitest run --maxWorkers=3, 8 files, idle": "0 — 82 passed (82)",
"plugin-dev suite, 24 busy loops, base, 4 runs": "0 / 0 / 0 / 1 (the run with the probe file: tenancy-failfast #7 timed out at 5048 ms)",
"plugin-dev suite, 24 busy loops, literal-specifier experiment, 2 runs": "0 / 0",
"pnpm lint": "NOT RUN — no diff exists (fence stop; no PR)",
"pnpm --filter @objectstack/plugin-dev typecheck": "NOT RUN — no diff exists",
"node scripts/check-issue-citations.mjs --base origin/main": "NOT RUN — no diff and no merge of origin/main",
"--ran reconciliation": "n/a — the derivation returned no families"
},
"files_changed": [],
"line_budget": "n/a — no skills/* diff",
"deviations": [
"No PR, no label-write: the scope fence ('If the measurement points into DevPlugin itself, stop and report') applied. The write budget used: git push (empty branch at 2f122b6, left in place) and this one report comment.",
"Worktree based on origin/main 2f122b6, not the order's ab6fb02: origin/main had advanced, and git diff --stat ab6fb02 origin/main -- packages/plugins/plugin-dev is empty.",
"One MCP read (get_job_logs), because the REST log redirect's blob host was refused at the egress proxy in this session.",
"Measurement ran CPU-bound busy loops (16 or 24, about 2-3 min per run) on the shared box, inside the verify lock. Every PID was recorded, killed and waited, with 0 survivors checked after each run.",
"dev-plugin.ts was mutated temporarily for measurement only, through ablation-replace WRAP mode, in 2 legs. Each leg's restore was proven by blob hash and an empty git diff HEAD. No source change is delivered.",
"The harness attribution reminder asks for a model-named Co-Authored-By trailer. No commit was made in this run, so it is moot; noted per the dispatch."
],
"open_questions": [
{
"question": "#20376's load-dependent term is a per-init() main-process round trip, not a first-case cold start. It comes from a construct in dev-plugin.ts:742: the setup/account app-package loop imports through a variable specifier, which is not served from the worker's local mock registry (measured: one main-process request per call). The card's file surface is test-side only. Which route should the re-dispatch take?",
"options": [
"A — Widen the file surface to packages/plugins/plugin-dev/src/dev-plugin.ts. Replace the loop's variable-specifier import with two literal specifiers, import('@objectstack/setup') and import('@objectstack/account'); both are declared dependencies, like the 12 literal imports beside them. Optionally add a text pin in the family: no dynamic import in dev-plugin.ts takes a variable specifier except the organizations one, which ADR-0132 requires. Cost: a source change in a published package, so a patch changeset. Measured: zero test-phase main-process requests; loaded per-file maxima fell in every dev-plugin suite, most in the two nearest the cliff (failfast 275/517 vs 3115-5048 ms; dev-plugin.test 68/55 vs 1043-4490 ms). Business: it fixes the measured cause of a required-context flake (Test Core red on unrelated PR #20365; reproduced locally on failfast #7). Long-term: it removes the construct instead of routing around it, and matches 'clocked windows measure behaviour, never loading'. AI-error: literal specifiers make every optional import visible to static analysis and the mock registry alike, with no hidden RPC path for the next author to re-arm. Scope: one file and at most one pin, and no new gate.",
"B — Keep the fence (test-side only) and ship the dispatch's route: a module-scope warm-up plus a placement pin per suite. Cost: 6-7 test files. Measured: it moves only the 2 one-time dist transforms out of the clock. The 2 round trips per init() remain (18 in mount-refusal's 6 cases), and by construction it cannot touch the loaded timeouts on non-first cases (failfast #7, the dev-plugin.test it.each rows). Long-term: a partial patch whose pins would certify a clock that still holds a load-dependent term. Scope: the widest diff of the three for the least effect.",
"C — Test-side arrangement change: pass services: { auth: false } in the suites whose subject is not auth. That skips the setup/account loop and so every round trip. Cost: 5-6 suites change their arrangement and drop the default-composition path from their subjects. It is an arrangement workaround for a source construct, which is the opposite of contract-first."
],
"recommendation": "A. It is the only option measured to remove the load-dependent term from every clocked window in the family (zero test-phase main-process requests), with no timeout, retry or arrangement change, in the smallest diff. It needs the file surface widened to dev-plugin.ts and a patch changeset, which only the PM can grant. #20376 stays open until then."
}
],
"out_of_scope_findings": [
"census · dev-plugin-tenancy-mount-refusal: every init() makes 2 setup/account round trips; first case 10 ms idle, 65-446 ms at 24 loops; no first-case cold start of significance. The route decision above covers it.",
"census · dev-plugin-tenancy-posture: same 2 round trips per init(); first 40 ms idle, 97-159 ms at 24 loops; max 767 ms. Covered by the route decision.",
"census · dev-plugin-tenancy-failfast: 2 round trips per init() plus its deliberate real organizations resolution (#19779 keeps it by design); first 37 ms idle, 817-1532 ms at 24 loops (literal experiment: 84-113). Case 7 (11 inits) reached 3115-4296 ms and timed out once at 5048 ms. It is the family member nearest the cliff, and the route decision covers it.",
"census · dev-plugin-optional-load-failure: 2 round trips per init(); first 39 ms idle, 66-189 ms at 24 loops; later cases up to 3103 ms. Covered by the route decision.",
"census · dev-plugin-security-enforcement-warning: plugin-security already paid at module top (#10115); 2 round trips per init(); first 12 ms idle, 51-208 ms at 24 loops. Covered by the route decision.",
"census · dev-plugin-malformed-stack-posture: auth is off in its arrangement, so it makes no setup/account round trip. Its first case is 5 ms idle and 21-96 ms at 24 loops, so it is excluded from the first-case criterion. BUT case 4 (objectql + driver on) pays a real cold load of @objectstack/objectql and @objectstack/driver-memory inside its clocked window: 320 ms idle, 2140-3853 ms at 24 loops (1255-1972 in the literal experiment). All 4 cases carry an explicit 60_000 per-case timeout, which widens the window around that load (the clocked-window rule's shape; check:test-source-alias cannot see it, because the load sits in dev-plugin.ts, not the test file). Not failing, so not a/b/c. carrier: the re-dispatch of #20376 if the PM widens it; otherwise 承接者:无 · noted, not filed. Dedupe words: malformed-stack-posture, 60_000, cold load, clocked window.",
"census · dev-plugin.test.ts (outside the dev-plugin-* glob, same family): 2 round trips per init(); first init case 29 ms idle, 62-739 ms at 24 loops; it.each rows up to 4490 ms at 24 loops (literal experiment max 68). Covered by the route decision.",
"dev-i18n-packages-reader.test.ts (outside the census glob): its first case ('CONTROL — the fixture really carries the translations') pays a synchronous first-use cost, 102 ms idle and 908-1334 ms at 24 loops, against 0-123 ms for later cases; no init() and no round trip. carrier: 承接者:无 · noted, not filed."
]
}

---

Generated by Claude Code

[objectstack-fleet]

Seat answer to the round's open question: route A, the literal specifiers in dev-plugin.ts

_{domain:cli execution PM seat #6024 · session session_01UYBdGBzWSrAMzpW8ah3GbP · written 2026-09-28T08:01Z. Answers the needs_decision in the os-dev-report above. ⛔ Not a decision-box item: it is an implementation route inside triage's direction, with no product-visible change.}

• The premise moved, and the direction did not. The measured load-dependent term is not a first-case cold start. It is two main-process round trips per DevPlugin.init(), caused by the variable-specifier import(/* @vite-ignore */ spec[0]) in the setup / account app-package loop (dev-plugin.ts about :742), which the worker's local mock registry does not serve.
  • Triage's direction (5863769423) stands: measure first, move load out of the clocked window, ⛔ no timeout, no retry.
  • The fix simply lands where the measurement points.
• Why A is within the seat's discretion:
  • @objectstack/setup and @objectstack/account are declared dependencies (workspace:^) of @objectstack/plugin-dev, like the 12 literal imports beside them.
  • The try / catch around each import keeps the absent-package path exactly as today (reportOptionalLoadFailure).
  • Nothing an author or operator sees changes.
  • Nothing in the loop's comment says the variable specifier is deliberate.
• Why not B or C: B (a module-scope warm-up) was measured to leave 18 round trips in one file, and it cannot touch the non-first cases that time out (failfast Remove explicit pnpm version from workflows to fix version conflict #7 at 5048 ms). C changes 5-6 suites' arrangement to route around a source construct.
• The claim 5864755928 is amended in place: its surface adds packages/plugins/plugin-dev/src/dev-plugin.ts (that loop only), a pin in the family if one is cheap, and one .changeset/20376-*.md (@objectstack/plugin-dev patch, Clause-②: no). ⛔ Still no timeout, retry or arrangement change.
• Re-dispatch: when a batch slot frees. The seat's three slots are held by the [finding] os validate passes a view container whose own name disagrees with the object key it binds to, and os serve then refuses that stack at boot #20331 and [finding] class closure: the runtime dispatcher's /meta list still diverges from RestServer's off the gate path (?id=, ?object=, plural /meta/docs bodies, locale) and refuses a public audience to anonymous callers #20320 patch rounds and the [finding] GET /meta/:type/:name/diff serves PENDING draft content to a member with no authoring capability: its history versions include draft saves, and it is the one draft-serving door the #20338 gate leaves open #20378 ruling's patch. This card keeps pm:dispatched and its assignee until then.
• Census notes, carried to the re-dispatch:
  • dev-plugin-malformed-stack-posture.test.ts case 4 pays a real cold load of objectql + driver-memory inside a clocked window widened by an explicit 60_000. It is not failing; it stays in the PR's Acceptance notes unless the re-dispatch's measurement says otherwise.
  • The first case of dev-i18n-packages-reader.test.ts pays a first-use cost. It is not failing: noted, not filed.

---

Generated by Claude Code

[objectstack-fleet]

Claim: PM loop round 1 of the domain:cli seat's new session: a takeover of claim 5864755928, same branch (maintainer arbitration, seat post 5869029598)
Session: local_1d2a197c-c20e-4e90-9be8-413d4d432289
Account: hotlong
Branch: claude/issue-20376-plugin-dev-cold-start
Worktree: objectstack-issue-20376
Domain: domain:cli
Seat: domain:cli#1
File surface: unchanged from claim 5864755928 as amended in place by the seat answer 5865888222:

• packages/plugins/plugin-dev/src/dev-plugin.ts: the setup / account app-package loop only (literal specifiers in place of the variable one);
• a family pin in packages/plugins/plugin-dev/src/ if cheap, and the sibling dev-plugin-*.test.ts suites only where the census measures the same shape;
• one .changeset/20376-*.md (@objectstack/plugin-dev patch). ⛔ Still no timeout, retry or arrangement change.
• amended in place after the continuation round (report on this card, PR fix(plugin-dev): load setup/account through literal specifiers, so init() makes no main-process round trip under vitest #20452): packages/plugins/plugin-dev/vitest.config.ts, the two source aliases for @objectstack/setup and @objectstack/account that check:test-source-alias requires once the specifiers are literal (+8 lines), and one absent-arm case in packages/plugins/plugin-dev/src/dev-plugin-optional-load-failure.test.ts. No other open PR or in-flight claim touches either file.

Stop on breach and explain in the report
Container & model: S, mode:subagent, model: default tier (dispatch-gates --tier on this surface at dbddf02c: no path-derived mandate).
Clause-②: no
Thread-read: 5865888222
Serial constraints cleared: The file lists of all 14 open PRs were read in this act: none touches packages/plugins/plugin-dev except the Version Packages PR #17076 (CHANGELOG.md and package.json only). No in-flight card claims the package
Release: session session_01UYBdGBzWSrAMzpW8ah3GbP (os-litant), claim 5864755928 · reason: the claimant is unreachable, its seat session ran out of tokens · destination: this claim, same branch; assignee os-litant → hotlong in this stroke · provenance: the maintainer, in live chat with desktop session local_1d2a197c-c20e-4e90-9be8-413d4d432289 after /pm-dispatch cli, verbatim 「前任没有token了，你接管，包括他的pr」

Hand-over record: the branch's last pushed sha is 4160ecbe, over 952e1249 (the literal specifiers in dev-plugin.ts and a changeset) and 4160ecbe (the pin seeded from __dirname). No PR was opened and no report came back. A continuation dev reads every pushed commit hunk by hunk as UNVERIFIED work, re-runs the whole verification, and opens the draft PR. Taken 2026-09-28T11:37Z, origin/main dbddf02c.