Escalation Policy
Who decides what, and when does a human need to get involved.
Priority Levels
| Level |
Name |
Who Can Decide |
Examples |
| P0 — Critical |
Alex (@root) mandatory |
Human only |
Production outages, security incidents, irreversible actions |
| P1 — High |
Agent + Alex notified |
Agent acts, Alex reviews |
Architecture changes, new dependencies, cost >$50 |
| P2 — Medium |
Agent autonomous |
Agent decides, Alex informed |
Routine tasks, documentation updates, minor refactors |
| P3 — Low |
Agent autonomous |
Agent decides, no notification |
Scorecards, journal entries, sweeps |
Escalation Triggers
Auto-escalate to Alex when:
- Issue labeled
priority:critical
- Issue in
workflow:blocked for >24 hours
- Meta-cortex fails to reach consensus in 3 iterations
- Two agents disagree on approach (conflict resolution needed)
- Action requires credentials Alex hasn't shared
- Cost estimate exceeds $50
How to Escalate
- Add label
priority:critical or workflow:blocked
- Comment with
@root mention explaining why
- Include context: what was tried, what failed, what's needed
Human Response Time
| Priority |
Expected Response |
| P0 |
ASAP (wakes human if needed) |
| P1 |
Within 4 hours during business hours |
| P2 |
Within 24 hours |
| P3 |
Best effort |
This policy prevents agents from making irreversible or expensive decisions without human oversight.