Video Proof

Record video proof of implemented features after coding tasks complete. Use when a coding agent finishes work and needs to visually verify and demonstrate th...

This skill appears to do what it says: it launches a server (via a start_command you provide), uses Playwright to record screen/screenshot/console logs, and writes proof artifacts. Before running: (1) review scripts/setup.sh because it will install node modules, download Chromium via npx playwright and may prompt for sudo to install ffmpeg — run it in a dev VM or container if you want isolation; (2) carefully review any proof-spec.yaml/start_command the agent will use — the skill will execute that shell command as-is, so do not let an untrusted agent choose a start_command or point it at production services; (3) avoid including secrets or production credentials in environment or repo paths used during recording, and review produced artifacts before committing (screenshots, logs, or API responses can include sensitive data); (4) if you need stricter control, run the scripts manually under an isolated user/container and inspect package.json changes and installed binaries first.

Type: OpenClaw Skill Name: video-proof-skill Version: 1.0.1 The skill is highly suspicious due to its design allowing arbitrary shell command execution and broad network access. Both `scripts/record-proof.js` and `scripts/api-proof.js` execute user-provided `start_command` values directly via `spawn('sh', ['-c', ...])`, which is a critical Remote Code Execution (RCE) vulnerability. The `SKILL.md` and `references/proof-spec.md` explicitly instruct an AI agent to generate this `start_command` based on its task, creating a direct prompt injection vector for RCE. Additionally, `scripts/setup.sh` uses `sudo` for dependency installation, and `scripts/api-proof.js` can make arbitrary HTTP/HTTPS requests, further increasing the attack surface.