**Verdict: no blockers.** The merge resolution keeps both sides' behaviour, and the new evidence supports the local claims. The one remaining gate is the push and CI: GitHub still shows head `0848428` with state `DIRTY`, so no required check has run on `1408a63` yet.

reviewed by claudeAgent/claude-opus-5-5 (the work was done on codex_pool/gpt-6.1-sol, so this is a cross-family review)

**What I checked**
- **Merge:** `1408a63` has two parents, `0848428` and `f7400ba`. `git show --remerge-diff` finds one conflict, the `restartObsidian` JSDoc at `tests/e2e/journey/actions.ts:367`.
- **Against #388:** the only differences from `f7400ba` are OE-ADOPT's intended ones. `launchObsidianInstance` plus `waitForInstanceReady` replace `relaunchIsolatedApp`, and the argv-only profile guard stays.
- **Against OE-ADOPT:** the only differences from `0848428` are #388's `beforeClose = ""` parameter, the IIFE close, and the JSDoc sentence. An empty `beforeClose` still produces valid JS.
- **No hidden conflicts:** #388's harness, verify-skill and fixture changes don't reference anything OE-ADOPT removed (no `XDG_RUNTIME_DIR`, `print-env`, `xvfb-run` or `Library/Logs` mentions). The lockfile still differs from master only in the root entry and `obsidian-e2e` (transcript item 368).
- **#388's journeys actually ran:** `quit-save` is gated by `test.runIf(OBSIDIAN_E2E_OBSIDIAN_HOME)`, and both logs show it passing rather than skipped. Both logs also show 53/53 tests, 11 files, the six sync-persistence cases, the four popout-save cases, and both restart journeys.
- **Every restart was observed:** each cleanup file records 4 process groups. That matches 1 launch plus the 3 `restartObsidian` callers (journey-playback, journey-resilience, quit-save).
- **Sequencing:** transcript item 361 shows one `&&` chain: E2E run 1, E2E run 2, lint, format, typecheck, build, unit tests, docs, size, then the dry-run stop. So nothing overlapped. The log contents match what the trail claims (1386 unit tests, 15 docs pages and 73 fragments, 314794 / 99368 bytes, `pids: none`).
- **Clean tree:** the working tree is clean and `.obsidian-e2e-artifacts` is git-ignored.

**Things to watch (none blocks the push)**
1. **The `xvfb-run.*` check in `run-suite.mjs` can't fail.** The runner (`withXvfb` in `node_modules/obsidian-e2e/dist/cli-BXyzH8Z-.mjs`) passes `-f <profile>/home/Xauthority` to xvfb-run. Per `xvfb-run` line 163, xvfb-run only creates a temp dir when no `-f` is given. The real cleanup proof is `profileRemoved` plus the process-group checks; Xvfb's argv contains the profile path, so it counts as an owned process. The check still guards against the old launch path coming back, but `resolution.md` and the PR body describe it as more than that.
2. **`cleanup-*.json` mixes measured and fixed values.** `survivingPids`, `remainingGroupMembers`, `matchingInstanceArgv` and `profileRemoved` are hard-coded and only written after the assertions pass. `recordedPids` and `recordedGroups` are measured. Also, `assert.equal(result.code, 0)` runs after the file is written. Both files record `code: 0`, so no current claim is wrong.
3. **The PR body is ahead of GitHub.** It describes merged head `1408a63` while GitHub still shows `0848428` as conflicting. This fixes itself on push. Don't report completion to the coordinator until the required checks are green on `1408a63`. Item 413 told the coordinator "local validation complete", which is accurate, but CI hasn't been confirmed.
4. **Trail gaps (no rows for these actions):**
   - Moving the old vault to `merge-f7400ba/previous-vault` (item 361) changed state.
   - Writing a new `run-suite.mjs` instead of reusing the original `process-proof.mjs`.
   - Uploading the evidence, editing the PR body, and the two interim coordinator messages (items 362 and 413).
   - Delegating this audit.

   Also, row 36 uses evidence paths relative to the log's own directory, while earlier rows use repo-relative paths. All the paths resolve.
5. **Typos in the trail and coordinator messages:** a space is missing before numbers and SHAs ("passed53", "across4", "as1408a63", "size314794"), in rows 32 and 35 and in items 362 and 413.

**Already-known follow-ups, unchanged by this merge**
- The macOS restart smoke check is still untested.
- Cleanup after a failure past spawn is only code-traced.
- Linux `print-env` drops `XDG_RUNTIME_DIR`, and the docs route CLI calls through npm.
- The local-file artwork bug is #394.
- The verify-skill walk was not re-run on the merged head. The PR body says so, and #388 only changed fixtures there.
- Heads-up: the gzip budget now has 632 bytes left. That comes from #388 on master, not from this PR.

I changed no files, launched nothing, and left no background children running.