The Autonomous Engines quest needs systematic triage of recurring engine failures to keep the autonomous discovery loop healthy.
## Steps
1. Query task_runs for recurring tasks with last_error not empty: `SELECT title, last_error, completion_notes, COUNT(*) as fail_count FROM tasks WHERE project_id IN (SELECT id FROM projects WHERE name='SciDEX') AND task_type='recurring' AND last_error != '' GROUP BY title, last_error, completion_notes ORDER BY fail_count DESC LIMIT 20`
2. For each failing engine: classify failure type (rate_limit, code_error, db_error, spec_mismatch)
3. For fixable code/db errors: create a one-shot fix task with the specific bug description and fix approach
4. For rate_limit errors: document the pattern and recommend frequency reduction
5. Write triage summary to spec
## Acceptance Criteria
- [ ] 20 recurring task failures triaged
- [ ] Fix tasks created for code/db errors
- [ ] Frequency reduction recommendations documented
- [ ] Triage summary committed