agentic-ai-engineering/skills/kits/multi-agent-orch/SKILL.md

59 lines
3.0 KiB
Markdown

# Multi-Agent Orchestration Kit
Scale from 1 agent to 20. Orchestration patterns that work.
## Included Skills
1. **Agent Team** — Dispatcher pattern: orchestrator delegates to specialists
2. **Agent Chain** — Pipeline pattern: YAML-defined sequential workflows
3. **P2P Coms** — Peer-to-peer agent communication (Unix sockets)
4. **P2P Coms-Net** — Networked P2P agent communication (HTTP/SSE)
5. **TillDone** — Task discipline with live progress
6. **Purpose Gate** — Session intent declaration
7. **Mental Model** — Agent-owned expertise files
8. **Domain Lock** — Per-agent filesystem permissions
## Architecture
```
┌─────────────────────────────────────────────────────────────┐
│ ORCHESTRATION PATTERNS │
├─────────────────────────────────────────────────────────────┤
│ │
│ Dispatcher (hub-and-spoke): Pipeline (sequential): │
│ ┌──────────┐ Step1 → Step2 → Step3 │
│ │ Orch │ Planner → Builder → Review │
│ └────┬─────┘ │
│ ├── Agent A P2P (flat): │
│ ├── Agent B Agent A ↔ Agent B │
│ └── Agent C Agent C ↔ Agent D │
│ │
├─────────────────────────────────────────────────────────────┤
│ AGENT INFRASTRUCTURE │
│ - Mental models compound across sessions │
│ - Domain locks prevent cross-contamination │
│ - TillDone ensures tasks are defined before work │
│ - Purpose Gate prevents aimless prompting │
└─────────────────────────────────────────────────────────────┘
```
## Quick Install
```bash
# Copy all skills
cp -r skills/* ~/.pi/skills/
cp extensions/*.ts ~/.pi/extensions/
# Configure teams
cp configs/teams.yaml ~/.pi/agents/teams.yaml
cp configs/agent-chain.yaml ~/.pi/agents/agent-chain.yaml
```
## Related Course Material
- M4 Multi-Agent Orchestration — full lessons on all patterns
- `lead-agents/` — depth-2 delegation system
- `ui-agents/` — 12-agent team system
- `ceo-agents/` — 8-agent board system
- `pi-vs-claude-code/` — all extensions with source