The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/detect_resources.py executes system management tools (nvidia-smi, amd-smi, sysctl, system_profiler) to query hardware state. These operations use subprocess.Popen with shell=False and hardcoded argument lists, preventing shell injection vulnerabilities.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes environmental data which could theoretically be influenced by an attacker. Ingestion points: Data is collected from environment variables and system command output in scripts/detect_resources.py. Boundary markers: The skill enforces a strict JSON schema for all outputs. Capability inventory: Resource probing and restricted local file writing via scripts/_common.py. Sanitization: String inputs are sanitized in scripts/detect_resources.py using _safe_text to remove non-printable characters, enforce length limits (128 characters), and redact sensitive identifiers like hostnames and absolute paths.\n- [DYNAMIC_EXECUTION]: A dynamic import of the psutil library is performed in scripts/detect_resources.py using a hardcoded string. This is a legitimate optional dependency for enhanced system telemetry.\n- [PRIVILEGE_ESCALATION]: File writing in scripts/_common.py implements significant defensive measures, including path restriction to the current directory, rejection of symbolic links, and enforcement of private file permissions (mode 0o600).

get-available-resources