Repository navigation
fix(omp): follow-ups from the integration-wait review (#2159) #2168
Description
Follow-up to #2159 (findings from the merge pass's review; none blocked).
- A project integration disabled in
config.ymlnow costs the full 30 s startup wait (Codex MEDIUM,server/omp.ts~155): main sent at once because it did not see project integrations; this PR does, but does not honourdisabledExtensions(also true for the agent-dir integration, as on main). Honour the effective disabled list before waiting. - An
-e/--extension/--hooknaming a directory or package is not recognized (Codex MEDIUM / omo LOW-1, ~151): only the basename is compared, so the send goes at once as before; resolve directories and packages the way omp discovers extensions. - The project check uses herdr's process cwd, not an absolute
--cwd(server/core/composer.ts~124); preferompLaunchDir(argv). - A named extension counts by basename without checking it exists (a missing relative path waits 30 s).
ompConfigDirkeeps the first of duplicated env variables whileompStoreLayoutrefuses;OMP_INTEGRATION_FILEis exported but only used inomp.ts.- A submit loop outlives a timed-out test in
server/core/omp-submit.test.tsand can break the next test'spane.getcount.
Activity
devswha commentedon Oct 28, 2026
Items from #1486 that neither #1825 nor #2159 fixed, tracked here so #1486 can close:
processEnvironis null there), so an integration found only throughOMP_PROFILE,PI_PROFILE,PI_CONFIG_DIRorPI_CODING_AGENT_DIRis not seen and that omp is sent to at once (server/core/composer.tsompProcess). Reading it through the kernel argument area, asroutingEnvironinserver/omp.tsdoes for routing words, would close it.pssynchronously (processStartedAt,server/process-start.ts).agent.promptstill reaches the paste-and-Enter fallback onagent_not_found. Every agent on main has the same narrow race.settings.jsonextensionsarray (agent or project level) are not seen (documented limit of fix(omp): wait for the integration wherever omp loads it, and not under --no-extensions (#1486) #2159).-epath with spaces reaches the process as several words, which omp joins back into one; we compare only the first word (documented limit of fix(omp): wait for the integration wherever omp loads it, and not under --no-extensions (#1486) #2159).devswha commentedon Oct 28, 2026
Closing under the new follow-up bar: a follow-up issue is kept only for a reproduced, user-visible MEDIUM finding. The remaining discovery gaps (disabled or explicit extensions, an absolute --cwd) are reasoned from code with no observed failed or delayed send; the macOS ps item was fixed by c6f77d7 (#1980). Below the follow-up bar: LOW/unreproduced; recorded on PR #2159 and in this issue for reference.