Shellwright generates video evidence to validate agent output

Dave Kerr introduced Shellwright, a tool that creates video recordings of shell sessions to serve as evidence in pull requests. Instead of relying solely on code review or passing unit tests, developers can use Shellwright to record terminal output from commands, such as running HTOP or interacting with Kubernetes via k9s. This allows reviewers to visually verify that an agent's changes work as intended without manually spinning up environments. Kerr compared this approach to using Playwright for web UIs, noting that large language models are effective at generating these recordings when prompted.

Kerr also discussed a workflow where agents are instructed to build Docker-based test suites for front-end applications, providing repeatable evidence of functionality. This method shifts the focus from reading code to validating outcomes, reducing the cognitive load on developers and increasing trust in agent-generated work. Kerr advocates for requiring agents to produce tangible evidence, such as video recordings or logs, to prove their work meets the desired outcome before human review.