# Area: Agent Operations

## Agent Specialization Map
Each agent owns ONE narrow domain to prevent context rot.

| Agent | Domain | Primary Function | Status |
|-------|--------|-----------------|--------|
| Alfred | Orchestration | Cross-agent coordination, daily ops, system health | Active (free tier) |
| Rick | Revenue | Sales pipeline, outreach, prospecting, deal closing | Idle (needs credits) |
| Devin | Delivery | Client deliverables, builds, deployments | Idle (needs credits) |
| Rene | Research | Market research, competitor analysis, vertical deep-dives | Idle (needs credits) |
| Daniel | Design | Visual assets, UI/UX, proposal graphics | Idle (needs credits) |
| Friedrich | Finance | Revenue tracking, cost analysis, invoicing | Idle (needs credits) |
| Laura | Legal | Compliance, contracts, terms of service | Idle (needs credits) |
| Persephany | People | Agent evaluation, team health, culture | Idle (needs credits) |

## Skill Incubator Pattern
From Kevin Badi (T2): analyze bottleneck → create skill → map to tools → iterate → deploy via cron

When developing new capabilities:
1. Identify manual bottleneck (what's slow or repetitive?)
2. Build skill to address it
3. Map skill to tools (APIs, CLIs, etc.)
4. Test and iterate until reliable
5. Deploy as cron job for autonomous execution
6. Monitor and improve

## Progressive Autonomy Rules
From Nat/Felix (T1): Don't give agents everything at once.
1. Start with read-only access to a domain
2. Prove reliable with small tasks
3. Expand to write access
4. Prove reliable with medium tasks
5. Expand to external-facing actions (with human approval)
6. Eventually: fully autonomous in their domain

## Build Repeatable Skills, Not Custom Projects [T2 — Kevin Badi]
"Agentic skills are eating SaaS." The future isn't standalone apps — it's skills that agents execute autonomously via cron.

**Operational rule:** When delivering any client service, always ask: "Can this become a repeatable skill?"

1. If you've done the same automation 2+ times → extract it into a skill file
2. Skills live in `~/.openclaw/skills/` and are mapped to specific agents
3. Skills are deployed via cron jobs for autonomous recurring execution
4. Skills are the product — custom one-off projects are the cost of learning what skills to build
5. Every client engagement should produce at least one reusable skill

**This is how we scale:** Build the skill once, deploy it to 50 clients. Don't rebuild from scratch for each.

## Session Management Rules
- **Never spawn codex sessions in TMP** — TMP gets cleaned out, kills long-running work
- **codex delegation pattern:** PRD → RALPH loop → codex executes → heartbeat monitors
- **Daily note tracking:** Every open session must be logged in daily note with path and status
- **Heartbeat monitoring:** Heartbeat checks daily note for open sessions — restart if died, report if finished