Designing Execution Policies

Designing Execution Policies That Don't Create Bottlenecks

An execution policy is the set of review and approval stages that a task must pass through before Paperclip marks it done. Get the policy right and you get quality control without slowing your agents to a crawl. Get it wrong in either direction — too many gates or too few — and you either become the bottleneck yourself or you ship work that should have been caught.

---

What an execution policy is

Every task in Paperclip has a lifecycle: open → inprogress → inreview → done. An execution policy defines what happens at the inreview stage. Without a policy, an agent completes its work and immediately marks the task done. With a policy, one or more conditions must be met before done is available: a human must approve it, automated tests must pass, the agent's manager must sign off, or some combination of these.

Policies live at the task level, not the agent level. You can apply a strict policy to high-consequence tasks and no policy to routine ones — even when the same agent handles both.

---

The problem with too many gates

The most common execution policy mistake is applying review stages because they feel responsible, not because they serve a specific purpose.

When every task requires human review before completion: