AgentSafety/ClawCare ? reverse-engineered prompt

Build me a Python tool called ClawCare that helps protect AI coding agents from malicious skills, plugins, and instructions.

I want it to work as a command line app that can scan a skill or plugin folder and flag risky things like command injection, credential theft, data exfiltration, privilege escalation, and prompt injection. It should support common agent setups like Claude Code, OpenCode, OpenClaw, Codex, and Cursor, and automatically figure out what kind of project it is. Also add a guard mode that can intercept shell commands while the agent is running, block dangerous ones, warn on suspicious ones, and keep a JSONL audit trail of what happened.

Please include a simple local HTML dashboard to view scan results and guard logs, plus a clean README style help output and basic tests. Make it feel like a real security tool someone could install with pip and use right away, and look up current docs online if you need to confirm the agent hook formats.