Every policy system has to answer one uncomfortable question: what happens to an action nobody wrote a rule for? Agents are creative, and they will eventually try something the policy authors did not anticipate.
Open by default hides gaps
If unmatched actions are allowed, a missing rule looks exactly like a safe action. The gap only shows up after the damage. The people who wrote the policy have no signal that anything was missed.
Closed by default makes gaps visible
In Aegisora an action without a matching policy is blocked. That can be inconvenient while you are setting up, and that is the point: the blocked action appears in the decision log, and you decide deliberately whether to allow it, escalate it or leave it blocked.
A practical way to start
- Run your agents in a test workspace and watch what gets blocked.
- Add narrow policies for the actions you expect.
- Use escalation for the actions that need a human.
- Review remaining blocks before moving to production.
The result is a policy set that describes what your agents are allowed to do, rather than a list of things somebody remembered to forbid.