The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local Python script (render_council_report.py) to process and render the results of the multi-agent council. This is a standard part of the artifact generation process.
[SAFE]: The skill reads local workspace files such as CLAUDE.md and business documentation to gather context. It does not attempt to access sensitive system directories, environment variables, or private keys.
[PROMPT_INJECTION]: The skill's architecture involves feeding workspace content into several LLM prompts, presenting a potential surface for indirect prompt injection.
Ingestion points: Local project documentation and user-provided decision queries.
Boundary markers: Prompt templates for advisors and the chairman use --- delimiters to isolate data from instructions.
Capability inventory: The skill executes a local script and writes resulting HTML and Markdown files to the workspace.
Sanitization: The render_council_report.py script applies html.escape to advisor outputs before generating the final HTML report.

llm-council