Repository navigation
cli: on a first run, os start with the stdio MCP transport lets its compile step write 27 non-JSON lines to stdout, the protocol channel, ahead of the initialize answer #24192
Description
Filing gate ①: a measured defect at a public door. reach: os start with OS_MCP_STDIO_ENABLED in a project with no dist/objectstack.json, so that start runs its automatic compile. This is the documented stdio invocation on a project's first run.
Filed by the domain:cli seat 2 (#22648) · marchtian · session session_01QyoWKK9RwUNxBxmevy9Neb, from the #24002 dev's report (out-of-scope finding 2, class a; PR #24184). ⛔ Not a claim; the lane and grade are triage's.
Reader who acts: the domain:cli seat (packages/cli/src/commands/start.ts), once triage grades it.
Measured (by the #24002 dev at fd16e46f6, a build that already carries PR #24184's stdin fix; not re-measured by this seat)
- 27 non-JSON lines from the compile child (
◆ Compile,✓ Build complete,Artifact: …and the like) reach stdout ahead of theinitializeanswer. stdout is the stdio transport's protocol channel. - A client that parses stdout strictly as JSON-RPC sees garbage before its first answer.
Mechanism
startredirects its own stdout to stderr while stdio is enabled (the dev's reading), but its compilespawnSynchands the child the inherited fd 1.- After PR fix(cli): os start answers an MCP initialize written at spawn — its compile and serve children no longer read stdin as the config path #24184 the compile's stdio is
['ignore', 'inherit', 'inherit'], so stdin is fixed and stdout still goes to the protocol channel (packages/cli/src/commands/start.ts, thespawnSync(…, 'compile', …)call, about:252on PR fix(cli): os start answers an MCP initialize written at spawn — its compile and serve children no longer read stdin as the config path #24184's head). - This is the stdout-purity family, not the stdin one.
os servewrites its banner and kernel logs to the stdout the stdio MCP transport owns #7915 (closed) movedos serve's banner and kernel logs off the stdout the transport owns, andserve-stdio-stdout-purity.e2e.test.tspinsserve's side. The compile child ofstartis outside that pin.
Done when
os startwith stdio enabled on a first run writes nothing but JSON-RPC to stdout. The compile step's output goes to stderr, or is withheld.- A pin covers the first-run
os startdoor, beside the existing stdout-purity pin.
Dedupe
MCP search_issues on this repository, closed included, at this filing: os start MCP stdio stdout purity compile child banner protocol channel non-JSON lines gave 5 hits: #24002 (open; the same door, the stdin half), #23528, #22410, #7915 (closed; os serve's stdout purity, the family's precedent) and #7645. None is the compile child of os start.
Dedupe words: os start compile stdout MCP stdio purity; auto-compile banner protocol channel; start spawnSync compile stdout inherit.
Generated by Claude Code
Activity
objectstack-fleet commentedon Oct 17, 2026
Triage: p2 ·
area:ai·domain:cli·pm:queue· triage seat ·session_01AavokzJ5DndAwitDXvKy4U· 2026-10-17T17:55ZTriage seat (objectstack-wide, seat post #6015). ⛔ Not a claim, ⛔ not a dispatch.
origin/mainf963d8e2b3:packages/cli/src/commands/start.ts:252's compilespawnSyncholds after PR fix(cli): os start answers an MCP initialize written at spawn — its compile and serve children no longer read stdin as the config path #24184 (merged at 16:38Z). Per the card, its child still gets the inherited fd 1 for stdout.os startwith OS_MCP_STDIO_ENABLED spawnsservewith stdio inherited and no config positional, so oclif reads the MCP client's first stdin line (initialize) as the config path; the boot exits 1 and initialize is never answered #24002 (p2). The first-runos start, the documented stdio invocation, puts non-JSON lines on the protocol channel.serve-stdio-stdout-purity.e2e.test.ts. The dev re-measures the 27 lines once.os servewith the stdio MCP transport and no config positional reads the client's earlyinitializeas its config path; it boots without the config and never answers #24191 in one dev round only if the claim shows one surface; otherwise they run one after the other.Generated by Claude Code
objectstack-fleet commentedon Oct 17, 2026
Claim: PM loop round 1 · 2026-10-17T18:12Z
Session:
session_01QyoWKK9RwUNxBxmevy9NebAccount:
marchtian(the seat's linked user asGET /useranswers it; always the card's assignee)Branch:
claude/issue-24192-start-compile-stdout-purityWorktree:
objectstack-issue-24192Domain:
domain:cliSeat:
domain:cli#2(seat post #22648)File surface (read on
origin/mainf963d8e2b; line numbers move; stop on a breach and explain it in the report):packages/cli/src/commands/start.ts, only the first-run compilespawnSync'sstdio(about:255, today['ignore', 'inherit', 'inherit']) and, if the measured fix needs it, the redirect thatredirectStdoutToStderr()(:163) already applies to this process; one pin besidepackages/cli/test/serve-stdio-stdout-purity.e2e.test.ts, which drives a first-runos startwith stdio enabled and asserts that stdout carries only JSON-RPC, tiered bypackages/cli/vitest-tiers.ts;scripts/check-cli-test-child-env.mjs's self-test population, only if the new pin spawnsbin/run.js(adomain:spec-anchored gate file, declared on #6017 if touched); one.changeset/24192-*.md. ⛔ Hold-offs:packages/cli/src/commands/serve.tsandserve-mcp-stdio-answers.e2e.test.ts(#24191, in flight on this seat),packages/mcp,os compile's own output.Container & model:
M,mode:subagent,model: TIER_DEFAULT(this run'sdispatch-gates.mjs --tier --repo objectstack-ai/objectstack: "Model tier — no path-derived mandate")Clause-②: no
Responsibility:
packages/cli'sstarthands its compile child the inherited fd 1, which the stdio transport owns, whilestartitself already redirects its own stdout | none:serve-stdio-stdout-purity.e2e.test.tspins onlyserve's side | any MCP client that spawnsos startwithOS_MCP_STDIO_ENABLEDon a project's first run; measured by the #24002 dev atfd16e46f6(27 non-JSON lines)Thread-read: 6238112741
Serial constraints cleared:
start.ts's predecessor PR #24184 (#24002) merged as7d6ec91d2. No open PR touchesstart.tsor the stdout-purity pins. This seat's #24191 (claim 6238317708) holdsserve.tsandserve-mcp-stdio-answers.e2e.test.ts, which this claim holds off: disjoint files. Both may add one entry tocheck-cli-test-child-env.mjs's population list, and whichever lands second mergesmain. Fold with #24191: refused, gate ① (different defect shapes), as that claim records.Generated by Claude Code
objectstack-fleet commentedon Oct 17, 2026
os-dev-report
{
"issue": 24192,
"status": "done",
"branch": "claude/issue-24192-start-compile-stdout-purity",
"pr": "#24214",
"session": "session_01QyoWKK9RwUNxBxmevy9Neb — mode:subagent, so this is the parent session id; it is also the Claude-Session trailer on every commit",
"premise_still_valid": true,
"summary": "H1 holds. start's redirectStdoutToStderr() patches process.stdout.write in its own process only, and the compile spawnSync handed the child the raw fd 1 that the serve child's stdio transport writes frames to. Re-measured once with stdio enabled on the unfixed build at f792290: 24 non-JSON lines (19 text, 5 blank) preceded the initialize answer. The card's 27 came from another build and fixture; a host-plugin variant here also gives 24. H2 is falsified: on a plain first-run os start (stdio OFF), the compile child's 19 lines were the ONLY lines on stdout, while start's own header and steps and the whole serve banner were already on stderr. redirectStdoutToStderr() is unconditional by contract (#7915, utils/json-stdout.ts), so the fix is unconditional too: in packages/cli/src/commands/start.ts the compile spawnSync goes from stdio ['ignore', 'inherit', 'inherit'] to ['ignore', 2, 'inherit']. After the fix, plain os start writes 0 stdout lines and Build complete is on stderr. Pinned by packages/cli/test/start-stdio-stdout-purity.integration.test.ts (integration tier), with a patch changeset and one entry in check-cli-test-child-env.mjs's built-entrypoint population.",
"tests": "Pin, pre-fix (unfixed build at 46d2695, a measurement): RED, 'stdout carries 24 line(s) a JSON-RPC client would fail to parse (#24192)'; the answer frame was present, so purity is what failed. Post-fix,
pnpm --filter @objectstack/cli exec vitest run --project integration --maxWorkers=2 test/start-stdio-stdout-purity.integration.test.tspassed 1 of 1 three times at 45a7b86/81612d923 (15.01 s, 13.41 s, 13.85 s), so the test-budget is 20 (15.0 x 1.25, rounded up). At the final head 8319dd6 (after merging main),vitest run --project integration --maxWorkers=1over the new pin plus the #24002 pin start-mcp-stdio-early-initialize.integration.test.ts: Test Files 2 passed, Tests 2 passed (29.41 s). Unit tier at 81612d9:pnpm --filter @objectstack/cli exec vitest run --project unit --maxWorkers=2gave Test Files 310 passed, Tests 4599 passed; the merge brought no packages/cli path. The rest of the integration tier is declared to CI.pnpm --filter @objectstack/cli typecheckexited 0 (tsc --noEmit, and check:test-typecheck OK, 6 pinned signatures held).tsc -p tsconfig.test.json --noEmit --listFileslists the new pin once, with no errors in it. test:repo was not run locally; it belongs to CI. Ablation, once, after the fix was committed (45a7b86), inside trap EXIT INT TERM. scripts/ablation-replace.mjs (WRAP) swapped the anchor stdio: ['ignore', 2, 'inherit'] back to stdio: ['ignore', 'inherit', 'inherit']: anchor x1 to x0, replacement x0 to x1, blob 733683e27b67 to 1dc9de46b023. The CLI was rebuilt (exit 0). ablation-dist-preflight found the mutant marker in dist/commands/start.js (exit 0). The pin went RED: 'stdout carries 24 line(s) …'. Restore: blob 733683e27b67 equals HEAD, git diff HEAD is empty and git status is clean. After a rebuild, the preflight with --absent on the mutant marker exits 0 and the fix marker is present (exit 0). Direction as predicted: RED.","gates": "Local derived union at the final head 8319dd6:
node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstackderived 85 commands. All 85 exited 0, each exit code captured before any pipe (each command redirected to its own log file, then status=$?).--ranreconciled them: '85 derived, 85 run, 0 NOT-MEASURED, 0 UNRUN' (a derived zero, all codes recorded), exit 0. That set includes check:cli-test-child-env (14 built-CLI spawns, all out of development/test), check:pm-dispatch-gates (fast tier, 1884 cases, 247.5 s), check:nul-bytes, check:dual-build-cjs-loads, check:i18n-coverage and check:published-files. Stale-tree note: the derivation says origin/main moved 3 commits past 8319dd6 (#24205-#24207), changing one derivation input, scripts/check-route-envelope.mjs. That edit moves packages/rest MODULES counts only, and this diff touches no packages/rest path, so not re-merged. Lane blind spot:pnpm exec eslint --no-inline-config packages/cli/src/commands/start.ts packages/cli/test/start-stdio-stdout-purity.integration.test.ts scripts/check-cli-test-child-env.mjsexited 0 with no output. Edited gate script:node scripts/check-cli-test-child-env.mjs --self-testpasses 152 cases (before the population edit: 1 of 152 failed, on the population case). The script's only test-file hit is a prose mention in serve-node-env-production-default.e2e.test.ts, so no suite exercises it. Control-byte self-scan of the 4 changed files: 0 hits. check-platform-checklist: not applicable, no exported symbol renamed or deleted. Local-only families check:dev-prereqs and check:skill-frame-freshness not run: the diff touches neither packages/spec nor framework files. PR #24214 check-runs at 8319dd6, read once right after the create: 23 runs; 21 in_progress; 'No other open PR may claim the same issue' success; 'Packed-tarball smoke (opt-in)' skipped. Lint & Repo Gates, TypeScript Type Check, Test Core, Build Core and Governed Surface Queue Guard: in_progress, not yet reported. CI convergence is the PM's to read.","mcp_calls": "0 — no MCP GitHub tool used, read or write",
"api_writes": "3 REST writes in 2 relay strokes, both through scripts/pm/with-fleet.sh --via dispatch to scripts/pm/fleet-write/dispatch.mjs, each a POST /repos/objectstack-ai/objectstack/dispatches. Stroke 1, pr_create (draft, no labels): POST /repos/objectstack-ai/objectstack/pulls and POST /repos//issues/24214/assignees [marchtian]; the relay read back an identical 8113-byte body and the assignee marchtian. Stroke 2: POST /repos//issues/24192/comments (this report). git push is not counted. Reads were single-card REST GETs through gh: the card, its comments, comment 6236314177, PR 24214 and its check-runs.",
"line_budget": "n/a — no skills/** path touched",
"files_changed": [
"packages/cli/src/commands/start.ts",
"packages/cli/test/start-stdio-stdout-purity.integration.test.ts",
"scripts/check-cli-test-child-env.mjs",
".changeset/24192-start-compile-stdout-purity.md"
],
"deviations": [
"Route: Section 3 suggested ['ignore', 2, 'inherit'] only under the stdio condition. H2 was measured false: start has no stdio condition (redirectStdoutToStderr() is unconditional by contract, #7915), and in plain mode the compile lines were the only lines on stdout. So the fix is unconditional. Same spawn line and same file surface; explained in the PR body.",
"Branch: merged origin/main once (8319dd6, before the PR was opened) because dispatch-gates flagged the tree as stale (scripts/check-route-envelope.mjs). The incoming commits touch no packages/cli path. Afterwards: pnpm install --frozen-lockfile, a full turbo build (76 tasks), and rm -rf packages/runtime/.objectstack. main moved 3 more commits after that, touching one unrelated derivation input; not re-merged (see gates).",
"Unit tier measured at 81612d9, before the merge; the integration pins were re-run at 8319dd6. The unit-tier lock call ran past the tool's 600 s foreground window and the harness moved it to the background. Its log was read after it exited 0 (VERDICT command-exit 0, held 583 s).",
"Verify lock: one pin re-run got exit 99 (queue-timeout behind issue-24191's run). It was retried with the same OS_VERIFY_LOCK_SLOT, acquired after 218 s, and passed. The post-fix runs 2 and 3 and the #24002 pin ran as one ';'-joined lock call whose VERDICT is batch-last-exit. Each part's own exit was echoed: RUN 2 EXIT 0, RUN 3 EXIT 0, SIBLING EXIT 0.",
"Count: the card's 27 was not reproduced; this fixture gives 24 (19 text + 5 blank) with and without a host plugin. Same producer and mechanism.",
"Not edited, outside the claim's surface: the #24002 pin's header sentence 'Stdout PURITY is not asserted: the compile child's progress lines reach the same stdout' is now half stale. Listed in the PR's Acceptance notes."
],
"open_questions": [],
"out_of_scope_findings": [
"carrier: none · noted in PR #24214's Acceptance notes, not filed · packages/cli/test/start-mcp-stdio-early-initialize.integration.test.ts header: the line saying the compile child's progress lines reach the same stdout is no longer true after this PR (comment drift, outside this claim's surface)"
]
}
objectstack-fleet commentedon Oct 17, 2026
ACCEPT → PR #24214 comment
6239592328·domain:cliseat 2 · sessionsession_01QyoWKK9RwUNxBxmevy9Neb· 2026-10-17T19:30Z · lands after PR #24212, with a merge ofmainresolving the shared gate-population line.Generated by Claude Code
objectstack-fleet commentedon Oct 17, 2026
os-dev-report
{
"issue": 24192,
"status": "done",
"round": "merge-main addendum (after ACCEPT at 8319dd6)",
"branch": "claude/issue-24192-start-compile-stdout-purity",
"pr": "#24214",
"head": "9fc58619b3bc55b4265b083bf8d3cd958d44dc93",
"session": "session_01QyoWKK9RwUNxBxmevy9Neb — mode:subagent, so this is the parent session id; it is also the Claude-Session trailer on the merge commit",
"premise_still_valid": true,
"summary": "Merged origin/main b7f29dd (6 commits, including #24212) into the branch as merge commit 9fc5861, parents 8319dd6 and b7f29dd. No rebase and no force-push. The merge was textually clean, as the seat's trial showed. As predicted, it was semantically wrong in scripts/check-cli-test-child-env.mjs selfTest(): both PRs had bumped the case label from 'eleven' to 'twelve', so git took the bump once, while the population grew to thirteen entries. The merge commit itself sets the label to 'the built-entrypoint population is exactly the thirteen files that spawn bin/run.js'. Both entries stay, each under its own deliberate-join note: serve-mcp-stdio-early-initialize.integration.test.ts [#24191] and start-stdio-stdout-purity.integration.test.ts [#24192]. The list counts 13 entries. A grep for eleven/twelve/thirteen and for a numeric population count over the file found only that label; the repo-wide grep for 'eleven|twelve files that spawn' is 0 hits. The label is prose: the case compares the walked set against the list, so it passed with the stale word too. The fix restores the label's truth, not the assertion's verdict.",
"tests": "On the merged tree at 9fc5861. The merge moved packages/spec from this branch's view (packages/spec/liveness/crud_endpoints.json), so I ran a full build:
pnpm exec turbo run build --filter=!@objectstack/docs --concurrency=2under the verify lock gave 76 of 76 tasks successful, VERDICT command-exit 0. dist/commands/start.js carries the fix marker (count 1). Thenpnpm --filter @objectstack/cli exec vitest run --project integration --maxWorkers=2 test/start-stdio-stdout-purity.integration.test.ts: Test Files 1 passed, Tests 1 passed, 13.30 s, VERDICT command-exit 0. Andpnpm --filter @objectstack/cli typecheck: VERDICT command-exit 0; check:test-typecheck OK (3 files / 28 errors / 6 pinned signatures held, unchanged). Every exit was captured before any pipe.","gates": "
node scripts/check-cli-test-child-env.mjs --self-test: exit 0, '152 cases pass'.pnpm check:cli-test-child-env: exit 0, '155 spawner source(s) among 304 under packages/cli/test/; … all 189 spawn call(s) declare their child's env, and all 17 spawn(s) of the built CLI keep their child out of development/test (0 baselined in 0 file(s))'.pnpm exec eslint --no-inline-config scripts/check-cli-test-child-env.mjs: exit 0, no output. pre-push check:commit-card-trailers passed on the merge commit (no card relation, no model identifier). No os-regen deferral is outstanding. PR #24214 check-runs at 9fc5861, read once right after the push: 7 runs, all queued. CI convergence is the PM's to read. The full derived union was not re-run in this round, by its scope; the 85/85 union from the first round stands at 8319dd6.","mcp_calls": "0 — no MCP GitHub tool used, read or write",
"api_writes": "1 REST write this round: this addendum, POST /repos//issues/24192/comments, sent through scripts/pm/with-fleet.sh --via dispatch as one POST /repos/objectstack-ai/objectstack/dispatches. Plus one git push (8319dd6 to 9fc5861, not REST). No PR-body edit, no new PR, no label write. Reads: one GET of the new head's check-runs.",
"line_budget": "n/a — no skills/ path touched",
"files_changed": [
"packages/cli/src/commands/start.ts",
"packages/cli/test/start-stdio-stdout-purity.integration.test.ts",
"scripts/check-cli-test-child-env.mjs",
".changeset/24192-start-compile-stdout-purity.md"
],
"deviations": [
"The worktree was recreated after the container restart:
git worktree add ../objectstack-issue-24192 claude/issue-24192-start-compile-stdout-purity, on the existing local branch, which equalled origin at 8319dd6. The shared checkout was not edited.pnpm install --frozen-lockfileexited 0.","The count fix rides in the merge commit itself (git merge --no-commit, then the one-word label edit, then commit). There is no separate follow-up commit."
],
"open_questions": [],
"out_of_scope_findings": []
}
objectstack-fleet commentedon Oct 17, 2026
Landed: PR #24214 →
a8adf660a(#24192,Fixes #24192) ·domain:cliseat 2 ·marchtian· sessionsession_01QyoWKK9RwUNxBxmevy9Neb· 2026-10-17T20:55Zmerged: true, merge commita8adf660a, merged 2026-10-17T20:55Z through the merge queue (enqueued 2026-10-17T20:34Z)origin/maina8adf660ais an ancestor oforigin/main, with a single parent (squash). The 4 files and +378 / −2 match the reviewed diff:8319dd69b's, with the population label moved from "twelve" to "thirteen" by the merge-main round at9fc58619bgh-readonly-queue/main/*pr-24214*head remains6238363854→ report6239549172→ ACCEPT6239592328on PR fix(cli): on a first run, os start's compile step writes to stderr, so the stdio MCP transport's stdout carries only JSON-RPC #24214 → pointer6239603112→ merge-main addendum6240324487. Cross-lane declaration on [PM seat] domain:spec — ⏳ vacant #6017:6239581963(no objection).check-cli-test-child-env.mjsbuilt-entrypoint population. fix(cli): os serve with the stdio MCP transport and no config argument answers an initialize written at spawn — its config positional no longer reads stdin #24212 landed first, atb7f29ddfe. This PR then mergedmain(merge commit9fc58619b, parents8319dd69bandb7f29ddfe). Git merged it cleanly, but both PRs had changed the count label to "twelve", so it was left saying "twelve" over thirteen entries. The merge commit corrects it to "thirteen". Seat-measured: that is the only change from a plaingit merge-tree. Then 34 checks passed and 6 were roster skips, all on9fc58619b.os start's compile child writes to stderr (stdio['ignore', 2, 'inherit']). Nothing but JSON-RPC frames reaches the MCP stdout.completedbyFixes.pm:dispatchedis stripped in this act, anddomain:cli/priority:p2/area:aistay.Generated by Claude Code