The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary function is to generate and execute shell commands across various environments (macOS/BSD, Linux/GNU, Windows/PowerShell) based on user requests.\n
Evidence: SKILL.md provides a workflow for parsing user input into CLI commands, emphasizing safe quoting for parameters.\n
Evidence: SKILL.md explicitly addresses destructive operations (modifying, deleting, etc.) by requiring confirmation and verification of tool help documentation.\n- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted natural language to drive system command execution.\n
Ingestion points: Natural language requests to create command lines (e.g., triggers in SKILL.md).\n
Boundary markers: The SKILL.md file defines explicit "Do not activate when" conditions to limit scope.\n
Capability inventory: Full CLI command generation and execution capabilities across all standard shell environments.\n
Sanitization: Instructions mandate the use of safe quoting for user-supplied arguments and tool verification to prevent unexpected command behavior.\n- [SAFE]: The profile management implementation in scripts/profile_store.py follows security best practices.\n
Evidence: The script includes a SENSITIVE_PARTS blacklist (including tokens, secrets, and passwords) to prevent credentials from being persisted in the shared profile.\n
Evidence: The script implements atomic file writes using tempfile and os.replace, and enforces restrictive file permissions (0o600) to protect user configuration files.

lov-help-cmd