The Agent Skills Directory

[SAFE]: The skill employs robust security practices for credential management. It resolves API keys for external providers (e.g., NVIDIA_API_KEY, CURSOR_API_KEY) directly from environment variables at runtime and explicitly instructs the coordinator never to log or expose these keys in transcripts or process arguments.
[COMMAND_EXECUTION]: The coordinator protocol executes local command-line tools such as codex, gemini, ollama, and cursor-agent to dispatch sub-tasks. It utilizes a security-conscious quoted heredoc pattern (COUNCIL_PROMPT_EOF) to pass prompt content, effectively preventing shell injection from potentially malicious user-supplied problem statements.
[EXTERNAL_DOWNLOADS]: The skill interacts with well-known and trusted technology services, including official GitHub APIs (via the gh CLI) and standard LLM provider endpoints (OpenAI, Google, NVIDIA NIM). These operations are documented as part of the core functionality for status detection and model inference.
[SAFE]: The installation script (install.sh) follows standard patterns for local deployment without using risky curl | bash installation methods. It performs local file operations to set up the skill environment in established user configuration directories.

council