superpowers:verification-before-completion

To stop an AI agent from claiming a task is complete without running tests, enforce an evidence-based verification gate. This requires the agent to execute specific testing or linting commands and validate the actual output before reporting success.

Evidence-based work completion in software development is a strict protocol requiring empirical proof from command execution. It mandates running verification commands, such as builds or test suites, and validating the output against expected results before asserting any task success.

To ensure code integrity when an AI fixes a bug, apply a rationalization prevention framework. This forces the agent to run the actual test command and confirm the output matches expected results, preventing false positives and unverified assumptions.

Yes, the verification before completion protocol works for linting and building workflows. It applies broadly across software development tasks by mandating that every success claim is backed by executing specific verification commands and checking their output.

Your AI assistant reports false positives because it relies on partial checks and assumptions instead of empirical evidence. Implementing a structured verification protocol forces the agent to execute actual commands and validate output before claiming a fix is successful.