You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
PRDs are currently treated as a task checklist — the agent marks items [x] when it thinks they're done, and the loop checks if all items are checked. There's no verification that the completed work actually satisfies the intent behind each task.
Idea
The PRD could become a structured contract that the verification step checks against:
Acceptance criteria: each task specifies observable outcomes ("endpoint returns 200 with valid auth", "form shows error on empty submit")
Edge cases: explicitly listed scenarios the implementation must handle
Architecture constraints: "use the existing UserRepo pattern", "no new dependencies"
Integration points: which existing tests must still pass, which APIs are affected
The verification step could then:
Parse acceptance criteria from the PRD
Check that test cases exist covering each criterion
Verify architecture constraints weren't violated (no new deps in package.json, imports follow patterns)
Run the full test suite to catch regressions
Context
This shifts the PRD from "list of things to do" to "list of things to prove were done." It makes the autonomous loop more trustworthy for complex tasks where "tests pass" isn't enough signal.
Closed as duplicate of #24 - PRD format (Phase 1) addresses this with BDD-style acceptance criteria, task verification steps, and done criteria per task.
Activity
dominicnunez commentedon Feb 21, 2026
Closed as duplicate of #24 - PRD format (Phase 1) addresses this with BDD-style acceptance criteria, task verification steps, and done criteria per task.