Acceptance criteria

An acceptance criterion is a testable statement of what “done” means for a task. In Masaro, it’s not a nice-to-have field; it’s the contract the whole agent workflow runs on.

Why a contract

When you hand work to an AI agent, you’re delegating decisions. Without a criterion, the agent fills the gaps with plausible guesses, and you’re left inspecting the output with no standard to inspect it against. You can’t verify work you never specified; you can only vibe-check it.

A criterion changes both ends of the hand-off:

  • Before the work: the agent knows the target. Ambiguity gets resolved in the task, where it’s cheap, not in the code, where it’s expensive.
  • After the work: you grade output against explicit checks. Met or not met. The review takes minutes and produces a real answer.

What a good criterion looks like

Verifiable, specific, scoped to the task. Compare:

WeakVerifiable
Add searchSearching for a task by a word in its title returns it in the results within 1s; an empty query shows recent tasks, not an error.
Improve error handlingA failed save shows an inline error, keeps the user’s input, and logs the failure with a request ID
Set up CIpnpm run verify runs on every push and a red run blocks merging

The pattern: an observable behavior, a boundary, and a failure mode. If you can’t imagine the check failing, it isn’t a criterion; it’s a wish.

Criteria in the agent loop

Tasks can carry multiple criteria. When an agent completes a task over MCP, it doesn’t just say “done”: it submits a summary and maps each criterion to whether it was met. That mapping is what you review. A task with unmet criteria isn’t done, no matter how much code got written.

A task in review, showing its description above two acceptance criteria: the first checked off and struck through as met, the second still unchecked.
Met or not met, one line at a time. That checklist is what turns a hand-back into a review you can actually finish.
Masaro helps you write these. Tasks that come out of Brainstorm arrive with drafted criteria, Suggest drafts up to three whenever you ask, and a task an agent claims with none at all gets them proposed right then. Every one of those routes ends in a suggestion you accept or dismiss, and you always get final edit. It’s your contract.

When to skip them

Real answer: sometimes. A trivial chore (“bump the dependency”) doesn’t need ceremony. The rule of thumb: anything you’d hand to an agent or a teammate you won’t be pairing with, and anything you’d argue about at review time, gets a criterion. Masaro won’t block you either way; it just makes the gradeable path the easy path.

Next: Writing acceptance criteria · The MCP hand-off