Guidance for AI agents working in this repository.
This is the open reference for Fleet Engineering — governing populations of agents with accountability. It complements loop-engineering.
- Read docs/concepts.md and docs/accountability-test.md
- Check FLEET.md for active fleet posture
- Patterns must update patterns/registry.yaml
- F0–F3 = fleet maturity levels (not loop L0–L3)
- Claw = agent with fixed service credentials
- Assistant = agent acting on behalf of invoking user
- Every pattern doc includes: goal, scale, week-one mode, human gates, failure modes
- Add vendor-only patterns without a DIY equivalent note
- Enable unattended write automation without human gates documented
- Break registry schema without updating
scripts/validate-registry.mjs
| Path | Purpose |
|---|---|
patterns/ |
Production fleet patterns |
templates/ |
AGENT-MANIFEST, permissions, budget |
tools/fleet-audit/ |
Readiness scorer |
tools/fleet-init/ |
Scaffold CLI |
docs/five-concerns.md |
Core discipline framework |