SOTA Tracker (Claw)

Provides daily-updated, authoritative rankings and metadata of state-of-the-art AI models aggregated from leading sources via JSON, API, or local queries.

This project appears coherent and fits its stated purpose (daily scrapers, exports, REST API, optional MCP). Before installing or enabling automation: 1) Review the scripts that will write to your home (update_sota_claude_md.py, any recommended copies to ~/.claude or ~/.cyrus) so you know what they change. 2) If you run scrapers, do so in an isolated environment (virtualenv, container, or VM) because Playwright will download browser binaries and scrapers make external network requests. 3) Don’t expose the REST API publicly without adding authentication or firewall rules; docs note wildcard CORS and no built-in auth by default. 4) If you enable GitHub Actions on a fork, review the workflow to understand what it commits; enabling workflows gives the repo the ability to push commits in that fork's context (standard GitHub behavior). 5) If you want to integrate as an MCP server, understand that adding the .mcp.json entry and running server.py will make a local service available to agents — only enable if you trust the code and network exposure. 6) No secrets are requested by the skill, which is good; still run tests and read init_db.py / scrapers/run_all.py to confirm they only access the public sources you expect. If any of these points are unclear, run the project locally in a sandbox first or ask for a targeted code review of specific scripts (e.g., update_sota_claude_md.py, server.py, scrapers/run_all.py).

Type: OpenClaw Skill Name: sota-tracker-claw Version: 1.0.1 The OpenClaw AgentSkills skill bundle is designed to provide up-to-date State-of-the-Art (SOTA) AI model information to an AI agent. It uses web scraping (Playwright) from legitimate sources (LMArena, Artificial Analysis, HuggingFace, Civitai), maintains a local SQLite database, and offers a REST API and MCP server. While it employs high-risk capabilities such as browser automation for scraping and persistence via systemd/cron for daily updates (`update_agents_md.py`), these are clearly aligned with its stated purpose. The agent-facing instructions in `server.py` and the example `SKILL.md` in `SKILLS_VS_MCP.md` are forms of prompt engineering to guide the agent in correctly using the SOTA tools, not for harmful objectives. No evidence of intentional harmful behavior like data exfiltration to unauthorized endpoints, unauthorized remote control, or obfuscation for malicious purposes was found.