Jev Completion Review for AI Agents | AutoJev

Jev completion review checks whether an agent has actually satisfied the objective before it reports success. The preset compares the requested outcome, completed work, verification evidence and known gaps, then recommends completion, more verification or continued work.

The workflow question

Is the objective complete and verified, or is required work still missing?

This creates a separate decision boundary between doing the work and declaring it done. It is especially useful for long agent runs where a plausible summary can hide an unresolved requirement.

Example completion input

{
  "objective": "Resolve a customer's duplicate-charge report",
  "completed_work": ["Matched both charges to the same order"],
  "verification": ["Customer identity and charge records verified"],
  "known_gaps": ["Refund has not been approved or issued"]
}

An illustrative result shape is:

{
  "recommendation": "verify_more",
  "completion_probability": 0.36,
  "guidance": "Keep the task open until the refund is approved, issued and verified."
}

Actual answers and probabilities depend on the supplied state and configured Jev provider.

Good completion boundaries

• Before an agent sends a final answer to the user.
• Before a coding agent reports that a fix is complete.
• Before closing a support, research or operations ticket.
• Before a workflow advances to billing, deployment or publication.
• After implementation but before irreversible follow-up actions.

The review should complement deterministic tests, not replace them. Test results, database state, deployment health and required approvals remain authoritative signals.

Run the preset

Use POST /api/v1/decisions/completion or open the completion example in the Playground. The completion-reviewer Skill teaches an agent to call the check before claiming success.

Related Jev resources

• Jev AI for agents
• Configure Jev MCP
• Install the completion Skill
• All Jev use cases

Frequently asked questions

Does completion review run tests?

No. The caller supplies completed work and verification results. Run deterministic tests before asking Jev to evaluate whether the evidence satisfies the objective.

Can the agent ignore the recommendation?

That depends on your policy. AutoJev returns advisory output unless your workflow explicitly enforces a completion gate.

What belongs in known gaps?

Include unresolved acceptance criteria, missing approvals, failed checks, unverified external state and work intentionally deferred outside the objective.