OpenClaw Skills Weekly

OpenClaw Skills Weekly — tracks trending ClawHub skills, generates GitHubAwesome-style YouTube video scripts with two-track ranking (Movers + Rockets).

This skill appears to do exactly what it says: fetch public ClawHub data, record snapshots to a local SQLite DB, optionally fetch READMEs, compute rankings, and call Claude (Anthropic) to generate scripts. Before installing/run: 1) inspect requirements.txt to ensure no unexpected packages are installed; 2) keep your ANTHROPIC_API_KEY secret and provide it only if you want LLM script generation (without it the pipeline skips the LLM step); 3) set SKILLS_WEEKLY_DATA and HOST_OUTPUT_DIR if you want to control where the DB and reports are stored (the default can write to a host mount if present); 4) if you don't want social capture, run with --skip-x or omit XAI_API_KEY; 5) consider running in an isolated container or VM and run the pipeline in snapshot-only mode first to observe behavior before enabling network-intensive modes. If you want extra assurance, review script_generator.py to confirm the Anthropic client uses the expected API endpoints and that no unexpected remote endpoints are contacted.

Type: OpenClaw Skill Name: skills-weekly Version: 1.1.0 The skill is classified as suspicious primarily due to a critical prompt injection vulnerability identified in `SKILL.md`. The instructions for the agent to execute `python3 run_weekly.py` directly embed user-provided input (`${EPISODE_NUM}`) into a shell command, creating a clear remote code execution risk. Additionally, `run_weekly.py` uses `subprocess.run` to execute `docker cp` commands, copying the skill's database from a Docker container to the host. While intended for legitimate data retrieval, this is a powerful and sensitive capability that could be abused if the agent were compromised, further contributing to the suspicious classification.