Evidence-Driven Testing is an agent skill by Michael Shimeles that has the agent test your app by hand while it records the screen. Each test and check is burned into the video as a caption, so the recording is the proof.
What it does
- Records the screen with a bundled FFmpeg script on macOS, Linux, or Windows, and writes a video, a report, and a manifest.
- Labels each test and marks every check as passed, failed, or untested with a reason.
- Falls back to scripted Playwright captures and numbered screenshots when there is no screen to record.
- Uses measured numbers or output pairs for API, performance, and other non-UI changes.
- Captures the failure before a bug fix, so the PR shows both the old and new behavior.
When to use it
Use it when a change alters how the app behaves and a reviewer needs to see it working, not just read that tests passed.
How to get it
The skill is free on GitHub, in Michael Shimeles's rstack repo. Install it with npx skills add michaelshimeles/rstack -s evidence-driven-testing, or drop -s to get the whole stack. The recorder needs Python 3 and ffmpeg with libx264 and the ass filter, and the headless helper needs Node 20 or later. The agent drives the app with computer use, or with cua-driver when it has no computer-use tools.
more skills
Evidence-Driven Testing is an agent skill by Michael Shimeles that has the agent test your app by hand while it records the screen. Each test and check is burned into the video as a caption, so the recording is the proof.
What it does
- Records the screen with a bundled FFmpeg script on macOS, Linux, or Windows, and writes a video, a report, and a manifest.
- Labels each test and marks every check as passed, failed, or untested with a reason.
- Falls back to scripted Playwright captures and numbered screenshots when there is no screen to record.
- Uses measured numbers or output pairs for API, performance, and other non-UI changes.
- Captures the failure before a bug fix, so the PR shows both the old and new behavior.
When to use it
Use it when a change alters how the app behaves and a reviewer needs to see it working, not just read that tests passed.
How to get it
The skill is free on GitHub, in Michael Shimeles's rstack repo. Install it with npx skills add michaelshimeles/rstack -s evidence-driven-testing, or drop -s to get the whole stack. The recorder needs Python 3 and ffmpeg with libx264 and the ass filter, and the headless helper needs Node 20 or later. The agent drives the app with computer use, or with cua-driver when it has no computer-use tools.
more skills
Evidence-Driven Testing is an agent skill by Michael Shimeles that has the agent test your app by hand while it records the screen. Each test and check is burned into the video as a caption, so the recording is the proof.
What it does
- Records the screen with a bundled FFmpeg script on macOS, Linux, or Windows, and writes a video, a report, and a manifest.
- Labels each test and marks every check as passed, failed, or untested with a reason.
- Falls back to scripted Playwright captures and numbered screenshots when there is no screen to record.
- Uses measured numbers or output pairs for API, performance, and other non-UI changes.
- Captures the failure before a bug fix, so the PR shows both the old and new behavior.
When to use it
Use it when a change alters how the app behaves and a reviewer needs to see it working, not just read that tests passed.
How to get it
The skill is free on GitHub, in Michael Shimeles's rstack repo. Install it with npx skills add michaelshimeles/rstack -s evidence-driven-testing, or drop -s to get the whole stack. The recorder needs Python 3 and ffmpeg with libx264 and the ass filter, and the headless helper needs Node 20 or later. The agent drives the app with computer use, or with cua-driver when it has no computer-use tools.

