cli test harness: the development column still draws a released port — a lost race drifts the child and portDriftError() reds a correct tree (35 files; #24295 closed the production column)

Carries the development column of #24295's family; #24295 keeps the production column (PR #24337).

Filing gate ①, as tooling: a false red on a correct source, reproduced deterministically. reach: every Test Core run (PR heads, queue builds, the hourly battery) executes the files below; a red from this column on main is NOT MEASURED in this act. Filed by the domain:cli seat 1 (#6024) · zhuangjianguo · session session_01SYEJqDL3NXxrJRBVqaMu9T, from #24295's dev report 6250714154 (its open question, answered here). ⛔ Not a claim. Lane and grade are triage's.

Measured (by #24295's dev, on the built CLI and the tsx entry)

reservePort() / randomPort() (packages/cli/test/helpers/serve-process.ts) bind 0.0.0.0:0, read the port, close the listener and hand the number over; the port is unheld until the child's own listen().
A development-posture child auto-shifts instead of exiting when that port is taken: bin/run-dev.js sets NODE_ENV=development before argv is parsed, so every tsx child is in this column. Measured on a port held by the test: serve --dev booted on HELD+1 (M8); tsx serve and tsx start booted on HELD+1 (M9, M10).
The harness then reds the run on purpose: runServe() and the bespoke spawners compare the banner's port with the requested one and throw portDriftError() (the serve-port-readback.e2e.test.ts held-port arm pins exactly that). So a lost race in this column is a loud red on a correct tree, not a wrong answer (runServe() children auto-shift port silently — bin/run-dev.js pins NODE_ENV=development, so a lost race is a FALSE GREEN and the test then talks to whatever else holds the port #12525 made it loud).
Family at that read: 31 tsx-entry files (runServe() callers and bespoke tsx spawners, including the 7 package-install-local-* os start files), plus 4 bin/run.js --dev / os dev files that PR test(cli): production-posture spawns take a kernel-assigned port; a pin holds every bin/run.js spawner to it #24337's pin ledgers as OUTSIDE_THE_PRODUCTION_COLUMN.

Wanted (the shape is the lane's call)

The same closed mechanism PR test(cli): production-posture spawns take a kernel-assigned port; a pin holds every bin/run.js spawner to it #24337 introduces for the production column: the child takes KERNEL_ASSIGNED_PORT (--port 0) and the test reads the bound port back (readKernelAssignedPort()), so no number crosses a gap. runServe() returning the bound port would carry most of the 31 at once.
PR test(cli): production-posture spawns take a kernel-assigned port; a pin holds every bin/run.js spawner to it #24337's pin widens its population from "names bin/run.js" to "spawns the CLI", and the 4 ledger entries are deleted once their files convert.
⛔ No retry-to-green, skip, .only or quarantine. ⛔ No new gate (the pin is a test, as on cli test harness: reservePort()'s close-to-spawn race reached main — the hourly full run went red on start-mcp-stdio-early-initialize (port 45195 taken before os start bound it); 46 test files share the draw #24295).
Likely over 1,000 changed lines; split by file family if so.

Done when

Every file in the enumeration spawns on the closed mechanism, or is named exempt with its reason in the pin's ledger.
The pin turns red when any CLI spawn, development or production, draws a released port.

Dedupe

MCP search_issues (closed included), "cli test harness port drift development spawn randomPort runServe released port race red" and "PORT DRIFT portDriftError serve auto-shift test flake": no open card covers the development column. Closed precedents: runServe() children auto-shift port silently — bin/run-dev.js pins NODE_ENV=development, so a lost race is a FALSE GREEN and the test then talks to whatever else holds the port #12525 (runServe() children auto-shift silently; made loud), [finding] getAvailablePort returns a DIFFERENT port than requested and nothing says so — the drift is announced nowhere, and three test files have been silently talking past it #12543, [finding] os serve publishes the REQUESTED port, not the bound one — with --port 0 it announces a port nothing is listening on, and the code comment asserts the opposite #13062, [finding] serve-publishes-bound-port e2e races its own port pick on a shared CI runner — dropped PR #15073 from the merge queue (expected 40733, got 40734) #15273, [finding] scaffold-e2e-boot-probe.test.ts boots its stub server on an advisory pickFreePort(38700) inside the Linux ephemeral range — "accepts the server it booted itself" exits 1 under load and dropped PR #20506 from the merge queue #20516. Control: the first query returns cli test harness: reservePort()'s close-to-spawn race reached main — the hourly full run went red on start-mcp-stdio-early-initialize (port 45195 taken before os start bound it); 46 test files share the draw #24295 (open), so the search reaches the subject.

Dedupe words: development column released port drift; runServe KERNEL_ASSIGNED_PORT sweep; portDriftError red correct tree tsx spawn.

Generated by Claude Code

Triage: p2 · `tooling` · `domain:cli` · `pm:queue` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-18T15:00Z

Triage seat (objectstack-wide, seat post #6015). ⛔ Not a claim, ⛔ not a dispatch.

The split is the one cli test harness: reservePort()'s close-to-spawn race reached main — the hourly full run went red on start-mcp-stdio-early-initialize (port 45195 taken before os start bound it); 46 test files share the draw #24295's dev asked for. cli test harness: reservePort()'s close-to-spawn race reached main — the hourly full run went red on start-mcp-stdio-early-initialize (port 45195 taken before os start bound it); 46 test files share the draw #24295 keeps the production column (PR test(cli): production-posture spawns take a kernel-assigned port; a pin holds every bin/run.js spawner to it #24337), and this card carries the development column. The reservePort() / randomPort() shape is unchanged on origin/main.
Grade p2, the family's grade. A lost race here is a loud portDriftError(), not a wrong answer. But it reds correct trees in every Test Core run, and 「平台测试清单反复变红，肯定要修啊」.
Serial: after PR test(cli): production-posture spawns take a kernel-assigned port; a pin holds every bin/run.js spawner to it #24337 lands, because this card widens that PR's pin population. If the change is over 1,000 lines, split it by file family, as the card says. ⛔ No retry, skip or quarantine. ⛔ No new gate.

Generated by Claude Code

Claim: PM loop round 4 · domain:cli seat 1 · #24342 slice 1 · 2026-10-18T17:10Z
Session: session_01SYEJqDL3NXxrJRBVqaMu9T
Account: zhuangjianguo (the seat's linked user as GET /user answers it; always the card's assignee)
Branch: claude/issue-24342-s1-dev-column-port
Worktree: objectstack-issue-24342-s1
Domain: domain:cli
Seat: domain:cli#1 (seat post #6024)
Ruling-ref: triage 6251398054 ("Serial: after PR #24337 lands, because this card widens that PR's pin population. If the change is over 1,000 lines, split it by file family, as the card says. ⛔ No retry, skip or quarantine. ⛔ No new gate."). PR #24337 landed as e4bcced0b (#24295, landing 6251538556).
File surface (read on origin/main 6a49efb9c; line numbers move; stop on a breach and explain it in the report):

packages/cli/test/helpers/serve-process.ts: runServe() spawns on KERNEL_ASSIGNED_PORT and returns the bound port, so the development column's main draw closes in one place. The reservePort() / randomPort() docblocks say what may still draw.
The runServe() callers among the development-column files (git grep -l "runServe(" origin/main -- 'packages/cli/test/**' names 21 files at this read), moved onto the returned port.
packages/cli/test/released-port-production-spawn.pin.test.ts (cli test harness: reservePort()'s close-to-spawn race reached main — the hourly full run went red on start-mcp-stdio-early-initialize (port 45195 taken before os start bound it); 46 test files share the draw #24295's pin): the population widens to every file it can see spawn the CLI for the files this slice converts, and the ledger entries of converted files are deleted. A file not converted in this slice stays in a ledger entry with its reason and exact count, so the pin stays green slice by slice.
Slice plan: this slice is the runServe() family. The bespoke tsx spawners (the 7 package-install-local-* os start files and the rest) and the 4 bin/run.js --dev ledger files are slice 2, unless this slice stays under 1,000 changed lines with them in. The dev measures it and says which.
No changeset: packages/cli/test/** ships in no package (skip-changeset, as cli test harness: reservePort()'s close-to-spawn race reached main — the hourly full run went red on start-mcp-stdio-early-initialize (port 45195 taken before os start bound it); 46 test files share the draw #24295).
⛔ Hold-offs: packages/cli/src/**; scripts/**; .github/workflows/**; no retry, skip, .only or quarantine; no new gate.
Container & model: L, mode:subagent, model: default tier (dispatch-gates.mjs --tier --repo objectstack-ai/objectstack over the surface: "Model tier — no path-derived mandate")
Clause-②: no
Basis for the Clause-② line: test harness only. No published accept set, export or public surface moves.
Responsibility: n/a — not a defect card (tooling: a test-harness race)
Thread-read: 6251398054
Serial constraints cleared:
PR test(cli): production-posture spawns take a kernel-assigned port; a pin holds every bin/run.js spawner to it #24337 (the production column and the pin) landed as e4bcced0b.
Open PRs' file lists were read in this act: none touches packages/cli/test/**.
test budget (domain:cli): 119 test file(s) over 10 s of wall in the hourly full battery (Part of #23151) #23181 (the cli test budget, packages/cli/test/**) waits behind this card.
This seat's in-flight work is disjoint: security: any user who may import can send runAutomations: false and skip every trigger and metadata-bound hook, so a sales rep bypasses an approval gate and a closed-record freeze through POST /data/:object/import #24348 (P0; packages/rest import door, possibly packages/core's import runner, spec export.zod.ts), PR refactor(service-storage): the storage download door calls the one request-to-ExecutionContext composition, resolveRequestExecutionContext #24364 (service-storage, core, runtime conformance) and PR fix(runtime): six door strings say "this system" or drop the noun, not "this environment" (cloud#3002 ruling A) #24367 (runtime/src/domains, docs/qa).

Generated by Claude Code

os-dev-report
{
"issue": 24342,
"status": "done",
"branch": "claude/issue-24342-s1-dev-column-port",
"pr": "#24386",
"session": "session_01SYEJqDL3NXxrJRBVqaMu9T — subagent run; the parent's harness-stamped id (the dispatch's Claude-Session line)",
"premise_still_valid": true,
"summary": "Slice 1 of #24342, the runServe() family. runServe() (packages/cli/test/helpers/serve-process.ts) now appends --port KERNEL_ASSIGNED_PORT when its caller passes no --port, and returns the bound port as ServeRun.port, read back with readKernelAssignedPort(). A caller that passes its own --port keeps portContentionError() and portDriftError(); the held-port arm relies on that. The 10 files that call runServe() dropped all 21 draws; serve-package-registry-always-on's own --dev boot was converted too, so that whole file left the ledger. The pin's population widened from 'names bin/run.js' to 'spawns the CLI': either entry spelled, or CLI/runServe imported from the harness by name, alias or namespace. Its ledger, renamed RELEASED_DRAW_LEDGER, holds the 25 unconverted files with their exact counts (28 draws) and the reason 'slice 2 of #24342'; the pin is green. The whole card measured over 1,000 changed lines, so per triage 6251398054 the 21 bespoke tsx spawners and the 4 bin/run.js --dev files are slice 2 (handover_slice2), and the PR says Part of #24342. A4: for the converted family the drift comparison is replaced by the read-back. A complete banner must name a non-zero http://localhost port, or runServe() rejects; this is pinned by 3 cheap cases in serve-port-readback and proven on a real boot (below).",
"tests": "Head 98af344, after merging origin/main 5da2e88; closure rebuilt with turbo run build --filter='@objectstack/cli...' (61/61, VERDICT command-exit 0). Queue tier (vitest run --maxWorkers=2): released-port-production-spawn.pin 4/4 (project unit), serve-package-declared-capabilities 5/5. Nightly tier (OS_TEST_TIERS=nightly): serve-port-readback 16/16, serve-boot-diagnostics 2/2, serve-boot-diagnostics-noise-budget 3/3, serve-app-runtime-hooks 1/1, serve-cluster-driver-diagnosis 8/8, serve-no-artifact 5/5, serve-organizations-host-resolution 3/3, serve-organizations-mount-failure 4/4, serve-package-registry-always-on 5/5. Total 56/56 in 11 files ('FINAL-TESTS queue=0 nightlyA=0 nightlyB=0 head=98af344ee'); the same 56 passed on the pre-merge head 1886250. pnpm --filter @objectstack/cli typecheck: exit 0, test layer holds its 3-file/28-error ledger; tsc --listFiles -p tsconfig.test.json puts all 12 changed files in the program, with 0 errors in them (the 28 are data-model-rules 23, i18n-extract-action-description 4, i18n-declared-surface-gate 1). eslint --no-inline-config over the 12 files: exit 0. Tier classification (vitest-tiers predicate) of every changed test file is unchanged vs base 6a49efb; the pin stays unit. Reverse verification, one-time, not kept: (1) through scripts/ablation-replace.mjs WRAP mode, serve-boot-diagnostics' runServe(dir, ['--log-level','debug'], ...) was changed to ['--port', randomPort(), '--log-level','debug']. The pin went 'Tests 1 failed | 3 passed (4)', naming 'serve-boot-diagnostics.e2e.test.ts: 1 released-port draw(s) (line 112), ledger allows 0'. That file names no entry and imports no CLI, so only the new runServe leg of the population catches it. Restore: blob == HEAD e97518ea9587, git diff HEAD empty. (2) A4 probe, a scratch e2e test deleted afterwards (git status clean): runServe(dir, [], ...) resolved with port 44449; the same boot with OS_BASE_URL=https://app.example.com rejected with 'CANNOT READ BACK THE BOUND PORT of os serve (bin/run-dev.js, via runServe): it was asked for port 0 ...'. (3) The forced lost race: serve-port-readback's held-port arm (an explicit held port) is rejected with PORT DRIFT, the red the card describes, while the kernel-assigned positive control in the same run resolves. No dist-resolved subject was mutated: the pin reads source through the TypeScript AST, so no build or dist preflight applied.",
"gates": "Local subset on head 98af344 (final commit, after the merge). node scripts/pm/dispatch-gates.mjs --commands (no paths): 52 commands, change set 12 paths vs merge base 5da2e88, 499 changed lines. Every command ran with its exit code captured before any pipe; all 52 exited 0. --ran: 'Run reconciliation — 52 derived, 52 run, 0 NOT-MEASURED, 0 UNRUN.' (exit 0). The 11 families of the path derivation's 'Declared WIDE population' block (init-service-contract, live-db-isolation, meta-type-normalized, optional-error-sink, resume-authority-declared, route-envelope, runner-env-posture, settings-bind-window, startup-registry-verdict, verify-stand-in, wildcard-fallthrough): all exited 0. check:cli-test-child-env: '155 spawner source(s) among 305 under packages/cli/test/**; no new bulk process.env copy reaches a spawned child, all 189 spawn call(s) declare their child's env', exit 0. A first pass, taken before the closure build had finished, read check:dts-closure exit 1 (declaration files absent) and check:dual-build-cjs-loads exit 3 (PREREQUISITE NOT MET). After the build both read exit 0, there and in the final union. check-platform-checklist not run: no exported symbol was renamed or deleted (ServeRun only gained a field). PR #24386 check-runs at report time on 98af344: 41 total, 8 success, 10 skipped, 23 in_progress. Test Core 1-6/6, Build Core and Governed Surface Queue Guard were in_progress; Check Changeset had a success. Not waited on. The 9 nightly-tier files do not run in this PR's Test Core; their readings above are local.",
"line_budget": "This slice: 499 changed lines (+306 / -193, 12 files), from dispatch-gates. The whole card was measured over 1,000: converting 31 of the 35 files came to 788 changed lines before the pin and the helper docblocks were touched (the 21 bespoke files alone are 607). The widened pin is 269 lines here, about 25 of them the slice-2 ledger, and the last three bin/run.js --dev files with their local drift functions were still to go. Split per triage 6251398054.",
"mcp_calls": "0",
"api_writes": "2 relay strokes, both through scripts/pm as objectstack-fleet[bot]. (1) pr_create via scripts/pm/fleet-write/dispatch.mjs: POST /repos/objectstack-ai/objectstack/dispatches (HTTP 204), run 39156305942 conclusion success. It executed POST /repos/objectstack-ai/objectstack/pulls (draft #24386), POST /repos//issues/24386/assignees and POST /repos//issues/24386/labels. Read-back: 13094 bytes sent and stored identical, assignees zhuangjianguo; a REST read shows draft=true and labels=['skip-changeset']. (2) This os-dev-report comment via scripts/pm/post-stamped.mjs: POST /repos//issues/24342/comments. git push is not a REST write.",
"deviations": [
"Merged origin/main 5da2e88 into the branch before pr_create (AGENTS §10). The merge commit 98af344 carries the model-free trailer pair, amended before the push. pnpm install was re-run because the merge moved the lockfile, the closure was rebuilt, and every test and gate above was re-run on that head.",
"The runServe-family boundary is drawn by behaviour (files that call the helper), not by the 21-hit grep. 4 hits (package-install-local-handlers, serve-migrate-exit-schema-parity, serve-port-drift-notice, serve-process-child-env) name runServe only in prose and spawn CLI themselves, so they are slice 2; package-install-local-handlers also stays with its family of 7.",
"As ruled, the four old OUTSIDE_THE_PRODUCTION_COLUMN entries now carry the reason 'slice 2 of #24342'. The constant is renamed RELEASED_DRAW_LEDGER, because an entry no longer means 'outside the production column'.",
"The first pin draft spelled the call shape runServe + '(' inside string literals. vitest-tiers.ts reads that shape in code position, string literals included, and moved the pin from the unit to the integration tier. This was caught before the PR: the shape is now spelled in pieces, and every changed file's tier was measured equal to base.",
"The relay's pr_create read-back line listed assignees but not labels. A REST read of #24386 shows labels=['skip-changeset'].",
"The slice-2 conversions of the 21 bespoke files were written and measured (607 changed lines), then restored to base on this branch. That patch is not pushed and lives only in this session's scratchpad; handover_slice2 carries the recipe instead."
],
"files_changed": [
"packages/cli/test/helpers/serve-process.ts",
"packages/cli/test/released-port-production-spawn.pin.test.ts",
"packages/cli/test/serve-app-runtime-hooks.e2e.test.ts",
"packages/cli/test/serve-boot-diagnostics-noise-budget.e2e.test.ts",
"packages/cli/test/serve-boot-diagnostics.e2e.test.ts",
"packages/cli/test/serve-cluster-driver-diagnosis.e2e.test.ts",
"packages/cli/test/serve-no-artifact.e2e.test.ts",
"packages/cli/test/serve-organizations-host-resolution.e2e.test.ts",
"packages/cli/test/serve-organizations-mount-failure.e2e.test.ts",
"packages/cli/test/serve-package-declared-capabilities.test.ts",
"packages/cli/test/serve-package-registry-always-on.e2e.test.ts",
"packages/cli/test/serve-port-readback.e2e.test.ts"
],
"handover_slice2": [
"Bespoke boot helpers that settle on portDriftError(out, what, port) and name portContentionError on exit (15 files). Recipe: spawn '-p', KERNEL_ASSIGNED_PORT; settle on readKernelAssignedPort(out, what), so settle takes Error or number; drop portContentionError on exit; carry the port, or base, on the handle; build each boot's base URL from its own handle. Several reuse ONE drawn port across 2-8 boots: package-install-local-uninstall-cleanups has 8 boots on one port, package-restart-acceptance, -handlers, -jobs and -boot-steps have 2-3. Files and draws: artifact-flag-precedence.integration 1; package-install-local-boot-steps 1, -handlers 1, -hooks 1, -jobs 1, -jobs-pull 1, -jobs-shared-name 1, -uninstall-cleanups 1 (all .integration); package-restart-acceptance.integration 1; serve-config-boot-flat-docs.integration 1; serve-config-boot-residual.integration 2 (also bootRefused); serve-host-config-metadata-door.integration 2; serve-organization-record-page.e2e 1; serve-view-container-read.integration 1.",
"Their own banner read-back, no drift check: serve-runtime-assets-branding-warning.e2e 1 (reservePort + boundPortFromBanner, portContentionError on exit) and start-signal-forwarding.e2e 1 (reservePort; argvFor takes the port; readback via boundPortFromBanner). Recipe: pass KERNEL_ASSIGNED_PORT and read back with readKernelAssignedPort.",
"Copies of the runServe shape: serve-app-anchored-optional-import.e2e 1 (runServeFrom: portContention + portDrift. Read back only when the banner is complete, because the undeclared leg dies before listen(); the header's port-apparatus section needs rewriting). serve-process-child-env.e2e 1 (probeOrigin's default randomPort() becomes KERNEL_ASSIGNED_PORT; readyVerdict already handles a requested 0 through portDriftError's 0 branch). serve-migrate-exit-schema-parity.integration 1 (migrate-and-exit, a single '--port', randomPort() argument).",
"serve-port-drift-notice.e2e 1: only the NEGATIVE arm draws; boot(KERNEL_ASSIGNED_PORT) never shifts, because getAvailablePort(0) answers 0. The POSITIVE arm keeps its real held neighbour.",
"serve-publishes-bound-port.e2e 2: RE-LEDGER with the real reason, do not convert. reservePort() is the connect instrument's NEGATIVE arm, a port probed free and left unbound, with no spawn on it. randomPort() is the non-zero request that IS the arm's subject; that arm already tolerates the race by reading the child's own drift notice.",
"bin/run.js --dev / os dev (4): dev-boot-output-order.integration 1 (a pty command line '-p PORT'; it reads no port). serve-mcp-capability-collision.e2e 1, serve-mcp-stdio-answers.e2e 1, serve-stdio-stdout-purity.e2e 1: module-level 'port = randomPort()', a local boundPort() and a local portDriftError(), boot 1 waiting on 'Server is ready', then HTTP on port. Recipe: spawn KERNEL_ASSIGNED_PORT, keep the local boundPort() gate (resolve only once a port is named), set port from boot 1's read-back, and delete the local portDriftError function and its docblock. Expect about 40 changed lines each.",
"Once every entry is gone, delete SLICE_2 and the ledger rows. Keep the pin's 'every ledger entry still names a file in the population' test; it reds a stale row. Update the reservePort()/randomPort() docblocks to name instruments only (plus serve-publishes-bound-port if it is re-ledgered). The PR is then 'Fixes #24342'."
],
"open_questions": [],
"out_of_scope_findings": [
"class: a · reach: public door os serve --port 0. Measured on 98af344 with tsx serve --port 0 against export default {};: the ready banner's 'Boot diagnostics' block prints 'WARN Port 0 is in use, using port 40073 instead' at the default log level, and the same line at info, beside a banner whose API row names the real port 40073. Port 0 is a declared legal request (packages/cli/src/utils/port-contract.ts, MIN_PORT = 0, 'a REQUEST, not an error'), not a busy port. Every kernel-assigned boot now prints it: #24295's production column and this slice's runServe family. It reds nothing today, because the noise budget counts it as informational. · evidence: packages/plugins/plugin-hono-server/src/hono-plugin.ts, the kernel:listening hook: 'const port = this.options.port ?? 3000; ... if (actualPort !== port) ctx.logger.warn(Port ${port} is in use, using port ${actualPort} instead)' compares the request with the bound port · Seam: runtime:plugin-hono-server kernel:listening warn, consumer: the ready banner's Boot diagnostics block · dedupe words: Port 0 is in use using port instead; hono-plugin actualPort warn kernel-assigned port; misleading port warning os serve --port 0"
]
}

ACCEPT → PR #24386 comment 6253877227 (slice 1, head 98af344ee; lands through the queue once green). Filed from the review: #24387. Slice 2 is the 25 ledgered files in RELEASED_DRAW_LEDGER; the card returns to pm:queue after slice 1 merges.

Generated by Claude Code

Release: domain:cli seat 1 · zhuangjianguo · session session_01SYEJqDL3NXxrJRBVqaMu9T · 2026-10-18T19:47Z · cause: slice 1 landed (Part of #24342) · destination: pm:queue, unassigned, for slice 2

Landed: PR #24386 → `c76e84922` (#24342 slice 1, `Part of #24342`)

Reading	Result
PR #24386 object	`merged: true`, merge commit `c76e84922`, merged 2026-10-18T19:45Z through the merge queue (enqueued 2026-10-18T19:19Z)
`origin/main`	`c76e84922` is an ancestor of `origin/main`, with a single parent (squash). The 12 files and +306 / −193 match what was reviewed at `98af344ee`
queue branch	no `gh-readonly-queue/main/pr-24386` head remains

Review chain: claim 6252728803 → report 6253829714 → ACCEPT 6253877227 → pointer 6253884308.
What landed: runServe() spawns on KERNEL_ASSIGNED_PORT unless a caller passes --port, returns the port the child bound, and rejects a complete banner that names no non-zero http://localhost port. The 10 runServe() caller files draw no port. test(cli): production-posture spawns take a kernel-assigned port; a pin holds every bin/run.js spawner to it #24337's pin now covers every file that spawns the CLI, and RELEASED_DRAW_LEDGER holds the 25 unconverted files (28 draws), each with the reason slice 2 of #24342.
Filed from this slice: os serve --port 0 prints "WARN Port 0 is in use, using port N instead" at the default log level — a kernel-assigned port is a legal request, not a busy port (plugin-hono-server's kernel:listening compare) #24387 (os serve --port 0 warns "Port 0 is in use").
What this card keeps: slice 2, the 25 ledgered files. They are listed by group in PR test(cli): runServe() takes a kernel-assigned port and returns the bound one; the released-port pin widens to every CLI spawn #24386's body under "What slice 2 converts": the bespoke boot helpers, the banner read-backs, the runServe() copies, the drift notice's negative arm, the deliberate draws to re-ledger, and the bin/run.js --dev files. test budget (domain:cli): 119 test file(s) over 10 s of wall in the hourly full battery (Part of #23151) #23181 still waits behind the whole card.
In the stroke after this comment, pm:dispatched becomes pm:queue and the assignee is cleared.

Generated by Claude Code

Claim: PM loop round 4 · domain:cli seat 1 · #24342 slice 2 · 2026-10-18T19:50Z
Session: session_01SYEJqDL3NXxrJRBVqaMu9T
Account: zhuangjianguo (the seat's linked user as GET /user answers it; always the card's assignee)
Branch: claude/issue-24342-s2-bespoke-spawners
Worktree: objectstack-issue-24342-s2
Domain: domain:cli
Seat: domain:cli#1 (seat post #6024)
Ruling-ref: triage 6251398054 ("If the change is over 1,000 lines, split it by file family, as the card says. ⛔ No retry, skip or quarantine. ⛔ No new gate."). Slice 1 landed as c76e84922 (release 6254355113).
File surface (read on origin/main cf16a0d18; line numbers move; stop on a breach and explain it in the report):

The 25 files of RELEASED_DRAW_LEDGER in packages/cli/test/released-port-production-spawn.pin.test.ts, each converted to KERNEL_ASSIGNED_PORT with the bound port read back through readKernelAssignedPort() (or the file's existing read-back), as PR test(cli): runServe() takes a kernel-assigned port and returns the bound one; the released-port pin widens to every CLI spawn #24386's body groups them under "What slice 2 converts". A file whose bespoke drift function becomes dead loses it.
The pin: each converted file's ledger entry is deleted in the same commit. A draw that is deliberate stays ledgered with its real reason in place of slice 2 of #24342, for example serve-publishes-bound-port.e2e's free, unbound port for the connect instrument's negative arm. The SLICE_2 constant goes once no entry uses it.
packages/cli/test/helpers/serve-process.ts, only where a converted file needs a helper the harness already half-carries. The reservePort() / randomPort() docblocks are updated to say what may still draw.
serve-port-drift-notice.e2e's NEGATIVE boot keeps its subject. If --port 0 changes what the arm proves, or meets os serve --port 0 prints "WARN Port 0 is in use, using port N instead" at the default log level — a kernel-assigned port is a legal request, not a busy port (plugin-hono-server's kernel:listening compare) #24387's Port 0 is in use warning, it is ledgered with that reason, not converted.
No changeset: packages/cli/test/** ships in no package (skip-changeset, as slice 1).
Size: if the slice measures over 1,000 changed lines, it splits by the body's groups into slices 2 and 3, and the dev says which group went first.
⛔ Hold-offs: packages/cli/src/**; packages/plugins/plugin-hono-server/** (os serve --port 0 prints "WARN Port 0 is in use, using port N instead" at the default log level — a kernel-assigned port is a legal request, not a busy port (plugin-hono-server's kernel:listening compare) #24387 is unclaimed and is not this card); scripts/**; .github/workflows/**; no retry, skip, .only or quarantine; no new gate.
Container & model: L, mode:subagent, model: default tier (dispatch-gates.mjs --tier, slice 1's surface: "Model tier — no path-derived mandate")
Clause-②: no
Basis for the Clause-② line: test harness only. No published accept set, export or public surface moves.
Responsibility: n/a — not a defect card (tooling: a test-harness race)
Thread-read: 6254355113
Serial constraints cleared:
Slice 1 (PR test(cli): runServe() takes a kernel-assigned port and returns the bound one; the released-port pin widens to every CLI spawn #24386) landed as c76e84922.
Open PRs' file lists were read in this act (7 open PRs): none touches packages/cli/test/**.
No open pm:dispatched claim of another card names packages/cli/test/**. test budget (domain:cli): 119 test file(s) over 10 s of wall in the hourly full battery (Part of #23151) #23181 (the cli test budget, same files) stays in pm:queue behind this card.
This seat's in-flight work is disjoint: runtime: an operation: 'update' action with no declared params writes any wire key into the row — the ledger says an undeclared param is refused (ADR-0104 D2); the console posts the patch as params, so both halves change together #24268 (packages/runtime/src/action-execution.ts, domains/actions.ts) and PR fix(verify): the dashboard date-window family also proves sums and derived measures, and says why an average stays out (#15953 slice 8) #24385 (packages/verify, in the merge queue). Batch 2: this and runtime: an operation: 'update' action with no declared params writes any wire key into the row — the ledger says an undeclared param is refused (ADR-0104 D2); the console posts the patch as params, so both halves change together #24268.

Generated by Claude Code

test(cli): every bespoke CLI spawner takes a kernel-assigned port and…

os-dev-report
{
"issue": 24342,
"status": "done",
"branch": "claude/issue-24342-s2-bespoke-spawners",
"pr": "#24396",
"session": "session_01SYEJqDL3NXxrJRBVqaMu9T — subagent run; the parent's harness-stamped id (the dispatch's Claude-Session line)",
"premise_still_valid": true,
"summary": "Slice 2 of #24342, the card's last: 23 of the 25 files in RELEASED_DRAW_LEDGER (25 draws, all six of slice 1's groups) now spawn the CLI on KERNEL_ASSIGNED_PORT and read the bound port back. Each boot builds its base URL from its own read-back. The bespoke settle steps become readKernelAssignedPort() verdicts, and the dead local portDriftError() copies and portContentionError-on-exit calls are gone. serve-port-drift-notice's NEGATIVE arm was measured: a --port 0 boot prints #24387's 'WARN Port 0 is in use, using port 43037 instead', and the arm's subject is a boot that binds the non-zero port it asked for. It stays ledgered with that reason, as the claim ruled; serve-publishes-bound-port keeps its 2 draws with slice 1's named reasons. SLICE_2 is deleted, and the reservePort()/randomPort()/portDriftError() docblocks say what may still draw. At 921 changed lines (under 1,000) nothing was split, and no ledger entry carries a slice reason, so draft PR #24396 says Fixes #24342 (assignee zhuangjianguo, label skip-changeset, Clause-② no).",
"tests": "Head 4fb8873, base cf16a0d. Closure built with turbo run build --filter='@objectstack/cli^...' (Tasks 60 successful, 60 total; VERDICT command-exit 0), then pnpm --filter @objectstack/cli build (exit 0). Every run went through scripts/pm/os-verify-lock.sh with a slot, each file in its own tier, '--project integration --maxWorkers=2'. Nightly (OS_TEST_TIERS=nightly): anchored-optional-import, organization-record-page, process-child-env, runtime-assets-branding-warning, start-signal-forwarding 'Test Files 5 passed (5) | Tests 23 passed (23)'; mcp-capability-collision, mcp-stdio-answers, stdio-stdout-purity 'Test Files 3 passed (3) | Tests 5 passed (5)'. Queue: flat-docs, residual, host-config-metadata-door, migrate-exit-schema-parity, view-container-read, package-restart-acceptance 'Test Files 6 passed (6) | Tests 37 passed (37)'; artifact-flag-precedence, dev-boot-output-order '2 passed (2) | 13 passed (13)'; install-local boot-steps/handlers/hooks/jobs-pull '4 passed (4) | 40 passed (40)'; install-local jobs-shared-name/jobs/uninstall-cleanups '3 passed (3) | 37 passed (37)'. In all, 23 converted files, 155/155. The pin (project unit) 'Tests 4 passed (4)'. Tier check with vitest-tiers.ts's own predicate at base and at head: 'TIER-CHECK files=24 changed=0'. pnpm --filter @objectstack/cli typecheck: exit 0, 'check:test-typecheck: OK ... 3 file(s) / 28 error(s) ... held'. tsc --listFiles -p tsconfig.test.json: 25 of 25 changed files in the program; the 28 errors are data-model-rules 23, i18n-extract-action-description 4, i18n-declared-surface-gate 1, none in a changed file. eslint --no-inline-config over the 25 files: exit 0. Measurement for the drift-notice arm, a scratch probe under the lock: control on released port 36499 gave 'API: http://localhost:36499/', DRIFT_NOTICE null, no Port line; --port 0 gave 'API: http://localhost:43037/', DRIFT_NOTICE null, 'WARN Port 0 is in use, using port 43037 instead'. Reverse verification, one-time: through scripts/ablation-replace.mjs (WRAP), serve-migrate-exit-schema-parity's "'--port', KERNEL_ASSIGNED_PORT, ...extraArgs" became "'--port', randomPort(), ...extraArgs" (anchor 1 -> 0, blob 9d7ef446bad7 -> febdea4dc573). The pin went 'Tests 1 failed | 3 passed (4)', naming 'serve-migrate-exit-schema-parity.integration.test.ts: 1 released-port draw(s) (line 96), ledger allows 0'. Restore: 'blob after restore 9d7ef446bad7...' == HEAD, 'git diff HEAD empty'; git status --porcelain was empty afterwards, read separately. The pin reads source through the TypeScript AST, so no dist preflight applied.",
"gates": "Local subset on head 4fb8873. dispatch-gates --commands --repo objectstack-ai/objectstack: 52 commands, change set 25 paths vs merge base cf16a0d, 921 changed lines. Each ran with its exit code captured before any pipe. 51 exited 0 on the first pass. check:dual-build-cjs-loads first read 'PREREQUISITE NOT MET' (exit 3: 10 packages outside the CLI closure had no dist). Re-run once those dist trees existed, it read exit 0 ('111 published require entry point(s) across 69 package(s) load'), and the ran list records that rerun. --ran: 'Run reconciliation — 52 derived, 52 run, 0 NOT-MEASURED, 0 UNRUN.' (exit 0). The 11 declared-wide families (init-service-contract, live-db-isolation, meta-type-normalized, optional-error-sink, resume-authority-declared, route-envelope, runner-env-posture, settings-bind-window, startup-registry-verdict, verify-stand-in, wildcard-fallthrough) and the 3 roster gates whose roster sits under packages/ (authz-resolver, error-code-casing, filter-alias-parity): all exit 0. check:cli-test-child-env: '155 spawner source(s) among 305 under packages/cli/test/**; no new bulk process.env copy reaches a spawned child, all 189 spawn call(s) declare their child's env', exit 0. check:nul-bytes: 'no raw ASCII control bytes', exit 0. check-platform-checklist not run: no exported symbol was renamed or deleted (the deleted portDriftError copies and SLICE_2 were module-local). PR #24396 check-runs at report time on 4fb8873: 41 total, 18 completed (none failed: Check Changeset success, Check PR Size success, the claim and closing-keyword checks success), 23 in_progress, including Test Core 1-6/6, Build Core, Governed Surface Queue Guard and every Lint and Type Check lane. Not waited on.",
"mcp_calls": "0",
"api_writes": "2 relay strokes, both through scripts/pm as objectstack-fleet[bot]. (1) pr_create via scripts/pm/fleet-write/dispatch.mjs: POST /repos/objectstack-ai/objectstack/dispatches (HTTP 204), run 39163689389 conclusion success. It executed POST /repos/objectstack-ai/objectstack/pulls (draft #24396), POST /repos//issues/24396/assignees and POST /repos//issues/24396/labels. Read-back: 12380 bytes sent and stored identical, assignees zhuangjianguo. A REST read shows draft=true and labels=['skip-changeset']. (2) This os-dev-report comment via scripts/pm/post-stamped.mjs: POST /repos//issues/24342/comments. git push is not a REST write.",
"deviations": [
"Not merged with origin/main before pr_create. Main moved cf16a0d to 584b422 during the run. Its 3 commits touch packages/cli/src (organization ownership, verify.ts) and none of the 25 files here, and git merge-tree --write-tree HEAD origin/main exits 0. The dispatch says to merge only for os-dev.md's reasons (a conflict, a main-caused red, a dequeue), so CI's merge ref reads the joint tree. AGENTS §10's pre-PR pull-and-rerun was not done; that is the trade.",
"The commits carry AGENTS.md's model-free trailer pair (Claude-Session + 'Co-authored-by: Claude'), not the harness reminder's model-named Co-Authored-By line. AGENTS.md says the pre-push hook refuses a model identifier in that pair, and the reminder yields to the repo's own rule.",
"The test batches and the gate run ran as backgrounded scripts. Each test batch acquired the lock itself, with a stable slot; the waits were bounded foreground blocks on their summary files. One Monitor was armed on the batch summary and ended by itself at ALL-BATCHES-DONE. Nothing is left running.",
"check:dual-build-cjs-loads read exit 3 (PREREQUISITE NOT MET) on its first pass and exit 0 on a rerun minutes later on the same head. The ran list fed to --ran records the rerun. Which earlier command produced the missing dist trees was not measured.",
"test:repo (the vitest repo project) was not run: the diff touches none of its tests and the derivation does not name it. It belongs to CI's Test Core.",
"Per-file test counts are reported per batch (2-6 files each): vitest's default non-TTY reporter printed batch totals only."
],
"files_changed": [
"packages/cli/test/helpers/serve-process.ts (docblocks only)",
"packages/cli/test/released-port-production-spawn.pin.test.ts",
"packages/cli/test/artifact-flag-precedence.integration.test.ts",
"packages/cli/test/dev-boot-output-order.integration.test.ts",
"packages/cli/test/package-install-local-boot-steps.integration.test.ts",
"packages/cli/test/package-install-local-handlers.integration.test.ts",
"packages/cli/test/package-install-local-hooks.integration.test.ts",
"packages/cli/test/package-install-local-jobs-pull.integration.test.ts",
"packages/cli/test/package-install-local-jobs-shared-name.integration.test.ts",
"packages/cli/test/package-install-local-jobs.integration.test.ts",
"packages/cli/test/package-install-local-uninstall-cleanups.integration.test.ts",
"packages/cli/test/package-restart-acceptance.integration.test.ts",
"packages/cli/test/serve-app-anchored-optional-import.e2e.test.ts",
"packages/cli/test/serve-config-boot-flat-docs.integration.test.ts",
"packages/cli/test/serve-config-boot-residual.integration.test.ts",
"packages/cli/test/serve-host-config-metadata-door.integration.test.ts",
"packages/cli/test/serve-mcp-capability-collision.e2e.test.ts",
"packages/cli/test/serve-mcp-stdio-answers.e2e.test.ts",
"packages/cli/test/serve-migrate-exit-schema-parity.integration.test.ts",
"packages/cli/test/serve-organization-record-page.e2e.test.ts",
"packages/cli/test/serve-process-child-env.e2e.test.ts",
"packages/cli/test/serve-runtime-assets-branding-warning.e2e.test.ts",
"packages/cli/test/serve-stdio-stdout-purity.e2e.test.ts",
"packages/cli/test/serve-view-container-read.integration.test.ts",
"packages/cli/test/start-signal-forwarding.e2e.test.ts"
],
"handover": "None. No slice remains: the ledger holds only the two deliberate entries, each with its real reason. If #24387 lands, serve-port-drift-notice's negative arm still stays drawn, because its subject is a boot that binds the non-zero port it asked for, which --port 0 cannot ask. PR body: no section is stale at report time.",
"open_questions": [],
"out_of_scope_findings": [
"carrier: #24387 · noted, not filed · Every kernel-assigned boot now prints plugin-hono-server's 'WARN Port 0 is in use, using port N instead' wherever the log level admits a warn: the 23 files converted here, on top of slice 1's runServe family and #24295's production column. It reds nothing: all 155 tests pass. It is one of the two reasons serve-port-drift-notice's negative arm stays ledgered, measured on 4fb8873: --port 0 printed 'WARN Port 0 is in use, using port 43037 instead'. The other reason, that requested 0 is not 'the port it asked for', survives a #24387 repair, so the ledger entry should not be expected to go when #24387 lands."
]
}

ACCEPT → PR #24396 comment 6254845694 (slice 2, the last; head 4fb88734c; lands through the queue once green, and the card closes on the merge). Filed from the review: none (#24387 already carries the Port 0 warning).

Generated by Claude Code

Landed: PR #24396 → `db4618b5c` · `domain:cli` seat 1 · `zhuangjianguo` · session `session_01SYEJqDL3NXxrJRBVqaMu9T` · 2026-10-18T21:16Z

Reading	Result
PR #24396 object	`merged: true`, merge commit `db4618b5c`, merged 2026-10-18T21:15Z through the merge queue (enqueued 2026-10-18T20:54Z)
`origin/main`	`db4618b5c` is an ancestor of `origin/main`, with a single parent (squash). The 25 files and +349 / −572 match what was reviewed at `4fb88734c`
queue branch	no `gh-readonly-queue/main/pr-24396` head remains

Review chain:
- slice 1: PR test(cli): runServe() takes a kernel-assigned port and returns the bound one; the released-port pin widens to every CLI spawn #24386 → c76e84922 (release 6254355113);
- slice 2: claim 6254381649 → report 6254823229 → ACCEPT 6254845694 → pointer 6254851514.
What landed: 23 of the 25 ledgered files spawn the CLI on KERNEL_ASSIGNED_PORT and read the bound port back, one read-back per boot. RELEASED_DRAW_LEDGER keeps only two deliberate draws, each with its real reason: serve-port-drift-notice's non-zero-request arm, and serve-publishes-bound-port's connect instrument and non-zero request. No entry carries a slice reason, so the card's "Done when" holds.
Filed from this card: os serve --port 0 prints "WARN Port 0 is in use, using port N instead" at the default log level — a kernel-assigned port is a legal request, not a busy port (plugin-hono-server's kernel:listening compare) #24387 (the Port 0 is in use warning on every kernel-assigned boot). Its fix will rewrite the drift-notice ledger reason that quotes it.
Unblocked: test budget (domain:cli): 119 test file(s) over 10 s of wall in the hourly full battery (Part of #23151) #23181 (the cli test budget, the same files) was waiting behind this card.
State: the card closed completed on the merge. pm:dispatched is removed in the stroke after this comment.

Generated by Claude Code

Navigation Menu

Repository navigation

cli test harness: the development column still draws a released port — a lost race drifts the child and portDriftError() reds a correct tree (35 files; #24295 closed the production column) #24342

Measured (by #24295's dev, on the built CLI and the tsx entry)

Wanted (the shape is the lane's call)

Done when

Dedupe

Triage: p2 · `tooling` · `domain:cli` · `pm:queue` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-18T15:00Z

Landed: PR #24386 → `c76e84922` (#24342 slice 1, `Part of #24342`)

Landed: PR #24396 → `db4618b5c` · `domain:cli` seat 1 · `zhuangjianguo` · session `session_01SYEJqDL3NXxrJRBVqaMu9T` · 2026-10-18T21:16Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

cli test harness: the development column still draws a released port — a lost race drifts the child and portDriftError() reds a correct tree (35 files; #24295 closed the production column) #24342

Description

Measured (by #24295's dev, on the built CLI and the tsx entry)

Wanted (the shape is the lane's call)

Done when

Dedupe

Activity

objectstack-fleet commented on Oct 18, 2026

Triage: p2 · tooling · domain:cli · pm:queue · triage seat · session_01AavokzJ5DndAwitDXvKy4U · 2026-10-18T15:00Z

objectstack-fleet commented on Oct 18, 2026

objectstack-fleet commented on Oct 18, 2026

objectstack-fleet commented on Oct 18, 2026

objectstack-fleet commented on Oct 18, 2026

Landed: PR #24386 → c76e84922 (#24342 slice 1, Part of #24342)

objectstack-fleet commented on Oct 18, 2026

objectstack-fleet commented on Oct 18, 2026

objectstack-fleet commented on Oct 18, 2026

objectstack-fleet commented on Oct 18, 2026

Landed: PR #24396 → db4618b5c · domain:cli seat 1 · zhuangjianguo · session session_01SYEJqDL3NXxrJRBVqaMu9T · 2026-10-18T21:16Z

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Triage: p2 · `tooling` · `domain:cli` · `pm:queue` · triage seat · `session_01AavokzJ5DndAwitDXvKy4U` · 2026-10-18T15:00Z

Landed: PR #24386 → `c76e84922` (#24342 slice 1, `Part of #24342`)

Landed: PR #24396 → `db4618b5c` · `domain:cli` seat 1 · `zhuangjianguo` · session `session_01SYEJqDL3NXxrJRBVqaMu9T` · 2026-10-18T21:16Z