| name | manual-test |
| description | Manually exercise code beyond the automated tests — python -c for libraries, curl for APIs, browser automation with screenshots for web UIs — and document the results. Use after tests pass, before declaring work done. |
Agentic manual testing
Passing tests do not prove the feature works as intended. Exercise the real
thing, with the mechanism matching what the code is:
Mechanisms
- Python library code: run edge cases directly with
python -c "..."
(multiline strings are fine). No files needed.
- Compiled or scripted demos: write throwaway programs under
/tmp (never in
the repo), compile/run them there.
- JSON APIs / servers: start a dev server and explore the API with
curl —
try a variety of endpoints, valid and invalid inputs, not just the happy path.
- Web UIs: use browser automation. Either Playwright, or
uvx rodney --help first to learn the tool, then drive a real browser.
Take screenshots and actually look at them with your vision capability to
confirm the UI renders correctly — layout, menus, visual details.
Rules
- If manual testing finds a bug, fix it via
/red-green-tdd so the case ends up
covered by the permanent automated suite.
- Document the session with Showboat: run
uvx showboat --help, then create a
document under notes/ (e.g. notes/api-demo.md). Use showboat note for
commentary, showboat exec to record commands with their real output, and
showboat image for screenshots. exec exists so the document shows what
actually happened, not what you hoped happened — never hand-write output.