The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/evidence.py script executes system tools including ffmpeg, ffprobe, and ps to capture screen recordings, verify video integrity, and manage processes. The scripts/run-playwright.sh script executes npm, node, and the Playwright CLI.
[EXTERNAL_DOWNLOADS]: The skill downloads the Playwright library from the NPM registry and the Chromium browser binary from Playwright's official distribution servers as part of the headless testing path setup in scripts/run-playwright.sh.
[DYNAMIC_EXECUTION]: The skill dynamically generates SubStation Alpha (.ass) subtitle files in scripts/evidence.py to burn test annotations into the video. Additionally, scripts/run-playwright.sh executes arbitrary Playwright scripts provided as arguments to facilitate headless UI testing.
[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from user-defined test targets and live application UI content while possessing significant system capabilities.
Ingestion points: User-supplied 'Test targets' and 'PR / issue' parameters in SKILL.md; live application UI state captured during recording or Playwright execution.
Boundary markers: None identified; instructions do not require specific delimiters or safety warnings for processing user-supplied data.
Capability inventory: Subprocess execution and shell access in scripts/evidence.py and scripts/run-playwright.sh; network access for dependency downloads.
Sanitization: scripts/evidence.py uses list-based arguments for subprocess.run to mitigate command injection. scripts/run-playwright.sh uses npm ci with the --ignore-scripts flag to prevent execution of malicious lifecycle scripts.

evidence-driven-testing