The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes several shell commands for running UI tests, processing video files, and interacting with the GitHub CLI.
Evidence: Use of bun run proof:ui, ffmpeg, ffprobe, and gh pr comment for automation and reporting.
[EXTERNAL_DOWNLOADS]: The skill contains references to documentation and skill templates hosted on the openclaw GitHub repository.
Evidence: Links to github.com/openclaw/openclaw for adaptation guidelines and instruction references.
[INDIRECT_PROMPT_INJECTION]: The skill captures live browser UI content which serves as a potential vector for indirect instructions if the captured site is malicious.
Evidence:
Ingestion points: Capturing real browser sessions and UI states via Playwright and ClawHub.
Boundary markers: Not explicitly specified for prompt interpolation.
Capability inventory: Includes file system access (.artifacts/), command execution (ffmpeg), and network interaction via the GitHub API (gh CLI).
Sanitization: The instructions include a mandatory inspection phase where the agent must verify the content of all media for privacy, readable text, and framing before publishing to a PR.

proof-video