| name | verify |
| description | Verify that a code change actually does what it's supposed to by running the app and observing behavior. |
| agent | build |
| argument-hint | <what to verify, e.g. "the login fix" or "PR |
Verify that the change described in $ARGUMENTS works correctly by exercising it at runtime.
Phase 1 - Understand the Change
- Identify what was changed:
git diff HEAD~1 --stat or the specific files/PR referenced.
- Determine the expected behavior: read the changed code, commit messages, or PR description.
- Identify the entry point: how does a user or test trigger this code path?
Phase 2 - Set Up
- Ensure dependencies are installed and the project builds.
- If a dev server is needed, start it in the background.
- If browser tools are available (AX_CODE_EXPERIMENTAL_BROWSER_AGENT), prefer them for UI verification.
- Otherwise, use curl/httpie for API endpoints or the project's test runner for logic changes.
Phase 3 - Exercise
- Trigger the changed code path with representative inputs.
- For UI changes: navigate to the page, interact with the component, take a screenshot.
- For API changes: send requests with valid and edge-case payloads.
- For CLI changes: run the command with typical and boundary arguments.
- Capture observed output (screenshots, response bodies, logs).
Phase 4 - Report
Produce a structured verdict:
Status: PASS or FAIL
What was tested: the specific actions taken.
Expected vs Observed: for each test point, what should happen vs what did happen.
Evidence: screenshots, response snippets, or log excerpts.
Issues found: if FAIL, describe each issue with reproduction steps.
Constraints
- Do not fix issues found — report them only.
- Do not modify source code.
- Clean up any background processes started during verification.
- If the change cannot be exercised at runtime (e.g. pure type-level change), say so and verify via typecheck + tests instead.