[JoAdo07]

Summary

Parallel prefs.mjs record calls shared one fixed tmp name, so the loser's rename died with ENOENT, and read-modify-write races dropped keys (#5333). Tmp names now carry the pid (as telemetry.mjs does), and both read-modify-write paths (recordProjectTier, recordUserSighting) run under the existing withFileLock with a re-read inside — including creating the parent dir first, which the lock itself needs.

Test plan

• New skills/media-use/scripts/prefs-concurrency.test.mjs: the issue's exact 4-parallel repro (all exit 0, all 4 keys kept) + sequential accumulation. 2/2 pass; the parallel case fails on unpatched code (verified via stash).
• node --check clean. Repo checks (check-comment-citations, ratchet, package suites) could not run here — no bun on this machine and npm install isn't supported for workspace ops — CI will run them.

Closes #5333

[jrusso1020]

Thanks for picking up #5333. The fix works: in a probe of 25 trials × 8 parallel node prefs.mjs record processes, this head kept every key and every user sighting, while main lost keys in about half the trials.

#5366 fixes the same issue with the same withFileLock approach, and we're going with that one, so I'd suggest closing this one as a duplicate. Here's why #5366 is ahead as submitted:

• It patches both copies of prefs-store.mjs. This PR only changes skills/media-use/scripts/lib/prefs-store.mjs, so scripts/check-media-use-copy-parity.test.mjs fails ("standalone and CLI copies differ").
• It regenerates skills-manifest.json. Here the media-use hash is stale, so the "Skills: manifest in sync" check would fail.
• Its tests race separate processes behind a start barrier and fail on main every time. They also cover the user sighting ledger, re-reading after a held lock, and abandoned locks. With the fix reverted, prefs-concurrency.test.mjs failed on main in only 3 of 5 runs.

Smaller things that would also trip CI here: oxfmt wants prefs-store.mjs:144 wrapped and the two JSON.parse( blocks in the test collapsed. The comment at prefs-store.mjs:75 ("used to share…") is history narration, which the Comments check rejects.

Thanks again for the clear write-up and the repro.

— Rames

[JoAdo07]

Closing this in favor of #5366, which covers the same race with the copy-parity and manifest updates called out in review. Thanks for the detailed feedback and for validating the fix.