Safe Exec

What to check before installing: - Verify the git repository (https://github.com/OTTTTTO/safe-exec.git) content and commit history yourself rather than relying solely on the SKILL.md claims. - Inspect the main script (~/.openclaw/skills/safe-exec/safe-exec.sh) to confirm how it intercepts commands: is it a wrapper that agents must call, or does it install shell hooks or modify startup files? Look for any edits to ~/.bashrc, ~/.profile, /etc files, or creation of daemons. - Understand agent-mode auto-approval: avoid setting SAFE_EXEC_AUTO_CONFIRM or exposing OPENCLAW_AGENT_CALL to untrusted agents. If you need automation, restrict which agents can set these vars and review the audit log frequently. - Ensure jq (declared dependency) is installed and that the git source is what you expect; the registry metadata and SKILL.md differ on declared requirements—clarify this discrepancy with the author or maintainer. - Test in an isolated or VM environment first (try harmless commands and known dangerous patterns) to confirm behaviour and that audit logs are generated and complete. - If you need a true system-wide interceptor rather than a wrapper, do not assume this skill provides that; ask the maintainer for explicit technical details about how global monitoring (if any) is implemented. Given the inconsistencies and the potential for agents to auto-execute commands via auto-confirm switches, treat this package with caution and validate the exact runtime behavior before enabling it in production or on important systems.