Video Skill

Run the video-skill pipeline to convert narrated videos into structured step data and enriched timeline-ready outputs. Use when a user asks to process a vide...

This package appears coherent for its stated purpose, but review these before running: 1) Provider trust: the enrichment steps will send transcript text and base64-encoded frame images to whichever base_url you configure for 'reasoning' and 'vlm' — only point those at services you control or trust. 2) Model downloads & docker: bootstrap_models.sh will download large model files (requires HF CLI and an authenticated account) and the docker-compose file pulls images from GHCR — verify sources and run in a machine with sufficient disk/GPU or use an isolated VM/container. 3) Local commands & subprocesses: the tool invokes ffmpeg and subprocess.run (clip extraction); run on files you trust and consider limiting permissions/using a sandbox. 4) Config review: config.example.json leaves api_key_env null; if you populate api_key_env make sure env vars are set appropriately and contain only credentials for intended providers. 5) Minor docs inconsistency: SKILL.md says 'no repo clone required' but many commands expect a local repo — follow the README/INSTRUCTIONS for correct setup. If you need higher assurance, review the remaining truncated source files (settings and any network code) and run the pipeline in a disposable container before using with sensitive data.

Type: OpenClaw Skill Name: video-skill Version: 0.1.2 The OpenClaw AgentSkills bundle 'video-skill' appears benign. Its primary function is to process videos using AI models, involving file I/O, external API calls to configured endpoints, and local execution of `ffmpeg` for video/frame extraction. API keys are handled securely via environment variables. The `SKILL.md` and other documentation provide instructions for using the skill's CLI, without any evidence of prompt injection attempts designed to manipulate the AI agent into unauthorized actions. The use of `subprocess.run` with a list of arguments for `ffmpeg` commands in `src/video_skill_extractor/clips.py` and `src/video_skill_extractor/frames.py` mitigates common shell injection vulnerabilities, as arguments are passed directly to the executable rather than interpreted by a shell. No signs of data exfiltration, persistence, or other malicious behaviors were found.