Autonomy
This is one policy kind in the ADR-014 governance engine. Policies resolve most-specific-wins with deny-wins at a tie, across scopes that run agent → user/role → workspace → team → org. Rules of the same kind combine rather than replace, so what matters is which rules exist, not which policy row holds them.
What it does
An autonomy policy sets a floor on how much an agent may
do without a human. Agents declare an autonomy level in
Agent.md — supervised, trusted, or
autonomous — and the effective level is
min(declared, floor).
What each level means
| Level | Writes |
|---|---|
| supervised | Every write parks for approval. |
| trusted | Plain writes run un-parked; destructive ones still park. |
| autonomous | Writes run without parking — subject to every other policy. |
Explicit approval policies and an agent's own requires_approval
flag always park regardless of level — autonomy lowers friction, it never
overrides an explicit gate.
At promotion
Promoting a shared-ring agent whose declared autonomy exceeds the org floor is refused (a policy conflict), so an agent cannot gain autonomy simply by being shared more widely.
