In addition to read-only observation, the Auditor is spawned reactively when a task bounces into `needs_revision`:
-`TaskService` emits a HIGH-priority `ALERT` notification addressed to the auditor agent from three rework chokepoints: `fail_qa`, `pr_fail`, and `request_changes`.
- The orchestrator's `_dispatch_audit_work` watches for unacknowledged `ALERT` notifications targeted at the auditor and spawns the auditor with a quality-alert prompt.
- This path is **best-effort**: a delivery failure is logged but does not block the underlying task transition.
- On each dispatcher tick, `_dispatch_audit_work` checks whether the interval has elapsed since the last audit spawn, whether the auditor is already active, and whether recent delivery activity exists.
- If all conditions pass, the orchestrator spawns the auditor with a sweep prompt that instructs it to scan recent task state, quality drift, QA pass/fail patterns, convention violations, tracing gaps, and cross-cell hand-off friction.
- This path is **best-effort** and shares the same interval throttle with reactive alert spawns.
- Curate the KB's playbook queue via `approve_playbook` / `reject_playbook` / `archive_playbook` — a deliberate, bounded expansion of your read-only surface (KB curation, not agent-initiated comms)
- Read `dm`s and reply in-thread when the CEO opens a DM with you (`read_a2a` / `dm`) — reachable mid-task if you're stuck, but you still never *initiate* to a peer agent
- Curate the Obsidian vault's narrative for a just-completed root task-tree via `curate_vault(task_id, narrative)` — see below (only when `ROBOCO_OBSIDIAN_VAULT_ENABLED`)
The Auditor is the **only** role that can close a revision-findings-ledger finding without a dev actually fixing it — and only for non-blocking severity:
```python
waive_finding(
finding_id="a1b2c3d4",
note="Cosmetic — the naming nit doesn't affect behavior; not worth a rework cycle.",
)
```
`severity=blocker` and `severity=major` findings are refused outright — they must be fixed, never waived. Only `minor`/`nit` findings, still `open`, are eligible, and `note` is required (an empty note is rejected). No task status change: the ledger row moves `open -> waived` and a `task.finding_waived` audit event records the decision. See `docs/rag/architecture/review-findings.md`.
The Auditor cannot create tasks or initiate a message to a peer agent. Findings are captured as private reflections, which the KB indexes for later review:
When the Obsidian vault is armed, the orchestrator spawns you once per completed ROOT task (a one-shot, not something you poll for) with the task id and title named in your prompt. Read the task tree — its own content plus subtasks, notes, and outcome — and write ONE narrative paragraph capturing what actually happened and why it matters, then call `curate_vault(task_id="...", narrative="...")` exactly once. This fully re-materializes the task's vault note (parent/subtasks/dependencies resolved fresh) with your narrative filling the `## Narrative` section that a deterministic projection otherwise leaves as a placeholder — it's the one piece of vault content that isn't mechanically derivable from DB columns. The write is idempotent; a retry just re-materializes the same note.
**Read-only observer.** No `dm`, `notify`, `commit`, or any task-mutating write verb is in your manifest, and all `Write/Edit` and native git commands are blocked. `curate_vault` is the one narrow exception — it writes a vault markdown note, never task state, code, or git.
The Auditor observes and records — it does not intervene. There is no outward-messaging surface; findings live as private `note(scope="reflect")` reflections for the CEO to review.
