Execute system-level shell commands directly within the thundarr-gpu container for monitoring and remote orchestration tasks.

This skill gives the agent unrestricted ability to run shell commands inside the container — it can read files, environment variables, network endpoints, and any mounted host artifacts (SSH keys, docker socket, config files). Only install if you fully trust the skill and the agent's autonomy. Consider: 1) Disable autonomous invocation (set disable-model-invocation = true) so the agent cannot run commands without your approval; 2) Remove or isolate sensitive mounts/credentials from the container (SSH keys, /var/run/docker.sock, host config directories); 3) Limit network egress from the container or firewall outbound connections; 4) Prefer manual invocation: require explicit commands from a human rather than letting the model decide; 5) Audit container file system and logs after use; 6) If you need limited monitoring/orchestration, prefer more narrowly scoped tools that expose specific APIs rather than an open shell. If you are unsure, treat this skill as high-risk and avoid installing it.

Type: OpenClaw Skill Name: exec-local Version: 1.0.0 The skill provides a raw shell execution primitive via the 'exec' tool in skill.json and skill.yaml, allowing for arbitrary command execution within the container environment. While SKILL.md describes its purpose as a bridge for orchestration and monitoring, the lack of any command filtering or input sanitization makes it a high-risk capability that could be leveraged for unauthorized system access or persistence.