safe-exec
High-risk skill: SafeExec provides local command interception and approval but contains code that executes arbitrary commands (`eval "$command"`) and offers non-interactive bypasses (`OPENCLAW_AGENT_CALL`, `SAFE_EXEC_AUTO_CONFIRM`). It also performs external network calls (`https://api.github.com`, `https://www.clawhub.ai`) and instructs modifying `~/.openclaw/config.json` and files under `~/.openclaw/safe-exec/`.