---

name: healer description: Healer monitoring expertise - detection patterns, API endpoints, dual-model architecture, and remediation workflows. Use when monitoring Play workflows or debugging agent failures.

Healer Skill

Healer is the observability and self-healing layer for CTO Play workflows. It monitors pod logs via Loki, detects issues, and orchestrates remediations.

When to Use

• Monitoring Play workflow execution
• Debugging agent failures (pre-flight, runtime)
• Understanding detection patterns (A10, A11, A12)
• Checking session status

---

Healer API Endpoints

Check Active Sessions

Copycurl http://localhost:8083/api/v1/sessions/active | jq

---

Detection Patterns

Priority 1: Pre-Flight Failures (within 60s of agent start)

Priority 2: Runtime Failures

Priority 3: Lifecycle Issues

---

Dual-Model Architecture

Copy┌─────────────────────────────────────────────────────────────────────────────┐
│                        DUAL-MODEL HEALER ARCHITECTURE                        │
│                                                                              │
│   DATA SOURCES                                                              │
│   ├─ Loki (all pod logs)                                                    │
│   ├─ Kubernetes (CodeRuns, Pods, Events)                                    │
│   ├─ GitHub (PRs, comments, CI status)                                      │
│   └─ CTO Config (expected tools, agent settings)                            │
│                              │                                               │
│                              ▼                                               │
│   MODEL 1: EVALUATION AGENT                                                 │
│   ├─ Parses and comprehends ALL logs                                        │
│   ├─ Correlates events across agents                                        │
│   ├─ Identifies root cause                                                  │
│   └─ Creates GitHub Issue with analysis                                     │
│                              │                                               │
│                              ▼                                               │
│   MODEL 2: REMEDIATION AGENT                                                │
│   ├─ Reads the GitHub issue                                                 │
│   ├─ Implements the fix                                                     │
│   ├─ Creates PR with changes                                                │
│   └─ Marks issue resolved                                                   │
└─────────────────────────────────────────────────────────────────────────────┘

---

Session Notification Flow

CopyMCP play() call
    │
    ▼
POST /api/v1/session/start
    │
    └─ Payload: {
         play_id,
         repository,
         cto_config: { agents, tools },
         tasks: [...]
       }
    │
    ▼
Healer stores session with expected tools per agent
    │
    ▼
CodeRuns start with Healer already aware

---

Watch Logs

Pod Logs

Copy# Watch all CTO pods
kubectl logs -n cto -l app.kubernetes.io/part-of=cto -f --tail=100

# Watch specific agent CodeRun
kubectl logs -n cto -l app=coderun -f

Loki Query

Copy{namespace="cto"} |= "error" | json

---

Pre-Flight Checklist (Verify within 60s)

For every agent run, Healer verifies:

Prompts

• Agent type identified
• Role matches task
• Template loaded
• Language context set

MCP Tools (from CTO Config)

• CTO config loaded
• Remote tools accessible
• Local servers initialized
• Tools-server reachable

---

Escalation

When issues detected:

1. Evaluation Agent creates GitHub issue with root cause
2. Remediation Agent attempts fix (if automatable)
3. Discord notification for P0/P1 critical issues
4. Human escalation if remediation fails

---

Configuration

In cto-config.json:

Copy{
  "defaults": {
    "play": {
      "healerEndpoint": "http://localhost:8083"
    },
    "remediation": {
      "maxIterations": 3,
      "syncTimeoutSecs": 300
    }
  }
}

---

Reference Documentation

• docs/heal-play.md - Full Healer specification
• crates/healer/ - Healer implementation
• crates/healer/src/scanner.rs - Detection patterns