Senpi runOmoCommand timeout QA

Objective audit

Requirement	Artifact	Evidence
Fix the final local bun run test:senpi timeout	packages/omo-senpi/src/components/ulw-loop/ulw-loop.test-support.ts	Fake omo commands now execute an absolute Node binary instead of nesting Bun under the Bun test runner.
Preserve the exact runOmoCommand behavior	packages/omo-senpi/src/components/ulw-loop/runtime.test.ts	Existing stdout/cwd assertions remain and now assert the child runtime is not Bun.
Prove RED before the fix	red-green.txt	PR #6310 recorded the actual 20-second full-gate timeout; a stale old-style fixture contained the Bun wrapper but no cwd.txt; unchanged code deterministically reported Bun 1.3.14.
Prove deterministic completion	red-green.txt	100 repeated runtime-test executions passed; duration fell from 147.81 seconds to 74.78 seconds.
Prove full gate	verification.txt	bun run test:senpi passed twice consecutively, 323 tests each.
Prove matching command surface	manual-qa.txt	Real omo ulw-loop status --json covered missing-plan and valid-plan paths and left no process behind.
Pass the HEAVY reviewer gate	review.txt	Goal, quality, security, hands-on, and history/context reviewers all passed the final diff with no blockers.
New worktree, PR, merge, cleanup	PR evidence after delivery	Branch fix/senpi-run-command-timeout; task worktree is removed after merge.

Root cause

createTempOmoBin() used process.execPath. Under bun test, that value is the Bun executable, so every fake status command launched another Bun process through /bin/sh. The original full gate recorded the exact runtime test timing out after 20 seconds. A pre-existing fixture directory from that failure class contained the old Bun wrapper and runner source but no cwd.txt, showing the runner did not reach its first write. The unchanged helper's runtime-identity assertion then deterministically failed with Bun 1.3.14.

The fixture now resolves the installed Node executable once with node -p process.execPath and invokes that absolute binary. Shell-wrapper behavior is deliberately unchanged, isolating the fix to one causal toggle: Bun child startup versus Node child startup. Production runOmoCommand() behavior and deadlines are unchanged.

The timeout itself is load-sensitive: a fresh unchanged gate and 100 isolated reruns completed successfully. The evidence therefore does not claim a new deterministic timeout reproduction. It combines the actual prior timeout, the incomplete old fixture artifact, deterministic wrong-runtime RED, independent hands-on process inspection, and the two-times faster stress result.

Validation summary

• Focused runtime test: 5 pass, 0 fail.
• Complete ulw-loop component suite: 27 pass, 0 fail.
• OMO Senpi package typecheck: exit 0.
• Runtime stress: 500 pass across 100 reruns in 77.34 seconds.
• Full Senpi gate: two consecutive runs, 323 pass / 0 fail each.
• Manual missing-plan status: expected structured error, exit 1, no leftover process.
• Manual valid-plan status: create exit 0, status exit 0, temporary state removed, no leftover process.
• Five-lane reviewer gate: unconditional PASS across all lanes.

Cleanup

• Runtime helper removes every omo-senpi-ulw-loop-* directory in finally; the final 100-run stress check reported zero leftover fixture directories.
• Manual QA removed /tmp/omo-senpi-run-command-qa.KL6n7S.
• No matching omo ulw-loop status process remained.
• No production source or generated bundle change is included.
• Setup resolves Node before creating the fixture directory, so discovery failure cannot leak a new omo-senpi-ulw-loop-* directory.
• The eleven stale fixture directories found during review were removed; the final verification baseline is zero.