Fleet Maturity Model (F0–F3)

Phased rollout for fleet engineering. Do not skip F1.

F0 — Ad-hoc Population

Symptoms:

• Agents in Slack, IDE, and SaaS with no central list
• Shared API keys in chat
• "Who turned that on?" is a frequent question

Minimum exit criteria to F1:

• FLEET-STATE.md or registry file exists
• Each production-touching agent has an owner
• Accountability test attempted for one recent action

F1 — Cataloged

Posture: Know what exists; humans approve risky actions.

Artifacts:

• patterns/registry.yaml or team agents/registry.yaml
• AGENT-MANIFEST.yaml per agent
• Permissions doc (clone / run / edit)
• Shared inbox or issue queue for approvals

Week one: No unattended writes to production. Report and approve only.

Exit criteria to F2:

• 100% of active agents in registry
• Identity model documented (claw vs assistant)
• Audit path exists (traces or structured logs)
• fleet-audit score ≥ 40

F2 — Shared Fleet

Posture: Good agents spread across team; budgets enforced.

Artifacts:

• Fleet budget file with per-agent caps
• Cross-agent search / audit playbook
• Clone-and-fork policy for team variants
• Kill switch documented and tested once

Exit criteria to F3:

• Budget alerts fire before hard stop
• Inbox SLA for risky actions
• Incident runbook references accountability test
• fleet-audit score ≥ 65

F3 — Enterprise Fleet

Posture: Policy-as-code, compliance, SLOs.

Artifacts:

• Autonomy tiers by data class
• Attested rollback for agent configs
• Central observability with retention policy
• Board-reportable reliability KPIs

Not every team needs F3. Most engineering orgs stabilize at F2.

Mapping to loop maturity

Loop level	Fleet implication
L1 report-only loops	Safe to register at F1
L2 assisted loops	Require F1 inbox + identity
L3 unattended loops	Require F2 budgets + kill switch minimum

See loop-engineering checklist.