Claude CodeArticle
Simon Willison: capture proof of an agent’s work
Showboat records commands and output; Rodney adds browser interactions and screenshots.
- Task: Review a working demonstration instead of relying on an agent’s claim that tests passed.
- Stack: Claude Code + Showboat + Rodney. Showboat creates Markdown with executed commands, output, and images; Rodney controls the browser.
- Flow: Ask the agent to read the tools’ help, demonstrate the feature, and record the result. The source links command-line and browser-accessibility demonstrations.
- Failure mode: An agent can edit the Markdown directly and make invented output resemble captured execution. Inspect how the evidence was produced.
- Evidence: Linked demonstrations and Simon’s account; not rerun here.
Source and demonstrations · February 10, 2026.