Verify code changes with AI
Match each completion claim to a fresh check. Read the output, verify the exit status and report which requirements were tested before handing work over.
Verify code changes with AI before telling a colleague a fault has gone. A passing build says that the application compiled. It tells you little about whether a customer can now save an order without creating a second copy.
The verification-before-completion skill asks the agent to identify the evidence a particular claim needs, run the relevant check and read the result. It can also compare a delegated change with the requested behaviour. If the check never ran or covered a narrower case, the handover should say so.
What it does
- Choose a check for the precise behaviour being claimed
- Inspect the complete output and exit status
- Repeat the scenario that originally failed
- Check requirements alongside automated test results
- Separate confirmed outcomes from untested assumptions
Install and use
- Install the skill in a compatible agent with access to the project and permission to run its normal checks. Prepare any test environment and sample data the chosen verification needs.
- Ask, “Before calling the duplicate-order fix complete, run the original save scenario. Report the number of orders created and the outcome of the related test.”
- Read the evidence with the handover. Record the actual blocker if a dependency or test environment prevented verification.
FAQ
Source
Made by Jesse Vincent and Prime Radiant. Original skill: verification-before-completion, published on GitHub under the MIT licence.
View on GitHub