[ITSGJY]

Summary

Fix Runner Job admission for Node project_build so a build planned from a Project's declared npm, pnpm, yarn, or bun manager is admitted under the same canonical process/digest contract as the typed Build Plan.

• Reuse ProjectBuildPlan::is_valid() at the Runner Job admission boundary, aligning it with Core wire validation and the Runner-owned planner.
• Preserve exact command shape, invocation digest, current manifest/lockfile provenance, queued-source revalidation, and the existing Job/Runner authority model.
• Add a deterministic no-child regression for all four managers with both explicit adapter=node and adapter=auto, including typed wire decode, Job queue admission, and a manager substitution with a recomputed digest rejected by the source fence.

The previous admission validator rebuilt the canonical command with an npm-defaulting helper. A valid pnpm/yarn/bun plan thus passed Runner planning and Core validation but was rejected before Job enqueue. The Runner's current Project recipe remains authoritative for manager selection.

Verification

• RED (original main@7b35aafb): The new regression fails on pnpm/Node: Runner rejected its own canonical Node build plan.
• GREEN (commit ed5afe29): The identical regression passes, including all eight manager/adapter combinations.
• Focused Runner project_build suite: 25 passed, 0 failed, 1 ignored (real Node/npm fixture).
• Ran the existing ignored real Node/npm build fixture separately: 1 passed.
• cargo fmt --all -- --check and git diff --check passed.
• Independent read-only Codex review: no P1/P2 findings.
• Fork GitHub Actions CI on exact commit ed5afe29eaa2c86e11aaa3b906cb6d3b0959a2f2: Success — 13 jobs passed, 0 failed, 5 correctly skipped by path-based selection (run 39567710526). Linux Runner/Server/workspace, Windows Runner, macOS Core, Contracts, MCP conformance, and the CI test gate passed.

Scope

Only the Runner admission check and its adjacent regression tests change (2 files). No new model-facing tool, schema/capability, executable-selection rule, process fallback, or persistence/retry contract. This follows the existing project-operation lifecycle tracked in #962.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[ITSGJY]

CI follow-up for the Windows Runner lane on upstream run 39577268778:

• The Runner test binary reported 1,025 passed, 4 failed, 55 ignored. All four failures were pre-existing project_format_tests cases returning format_timeout (rather than their expected formatting result/error); the new Node project_build admission regression passed.
• On the same exact commit ed5afe29, fork run 39567710526 passed the Windows Runner lane and all four formatting cases on the same windows-2025-vs2026 runner image. The upstream Windows lane also took over 60 seconds in an unrelated Ruff real-process test that completed promptly on the fork, consistent with transient child-process slowness.
• This PR changes only Node Build Job admission and its regression fixture; it does not change project_format or the Windows workflow.

Could a maintainer rerun just the failed test-windows-runner job against the unchanged head SHA? The contributor account lacks upstream Actions rerun permission. I have left the reviewed patch unchanged so that a rerun can verify the exact candidate.

[swjnxyf]

Reviewed PR #1244 in the MSI main checkout at 6fa4637b9e95130b2816f92afd733a30ec135a55.

The admission change correctly reuses ProjectBuildPlan::is_valid() rather than the old npm-defaulting canonical helper. This accepts Runner-selected npm/pnpm/yarn/bun Node build plans without allowing caller argv or manager substitution: the exact process/invocation digest must match, and the existing project_build::fence still replans the current manifest/lockfile at Job admission/queue execution. No new execution or effect authority is introduced.

Independent Windows checks:

• Deterministic eight-way (4 managers x Node/Auto) Node Build Job admission regression: passed twice.
• cargo test -p webcodex-core --lib project_build --offline: 12 passed.
• Wider webcodex-runner project_build group: 20 passed, 1 ignored, with 1 unrelated python_artifact_observation_uses_existing_metadata_and_bounds 5-second timeout under load; the exact test passed unchanged on isolated rerun in 2.14 seconds. This existing Python observation path is untouched by the PR.
• cargo fmt --all -- --check and git diff --check: passed.
• Exact-head upstream CI run 39580760135: 18/18 jobs successful, including Windows Runner.

No blocking findings or corrective commits needed. Approved for squash merge.