The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/bootstrap.py file uses subprocess.run to automate the installation of the agent-council package and its dependencies (pydantic-ai, pyyaml) using pip or pipx from the local skill directory. This is an intended and standard setup mechanism for skill packages containing CLI tools.
[DATA_EXFILTRATION]: The skill manages API credentials through environment variables (AGENT_COUNCIL_API_KEY) and local .env files. Analysis of the network-bound logic confirms that credentials are only used for authorized LLM provider API calls and are not exfiltrated to unauthorized domains.
[SAFE]: The skill implements a robust verification mechanism (FACTUAL_CLAIM_GUARDRAIL) that instructs agents to avoid asserting verifiable external facts (like domain availability or pricing). Additionally, a post-synthesis verification step scans the final output for unsubstantiated claims, demonstrating a security-conscious design.
[SAFE]: File operations are limited to creating run-specific output directories in /tmp/agent-council/ for storing debate logs and synthesis reports, following standard application logging practices.

agent-council