The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied brainstorming questions and constraints, which are then passed to sub-agents as a 'brief'.
Ingestion points: User input captured via $ARGUMENTS in SKILL.md is written to .council/brief.md by the orchestrator.
Boundary markers: The sub-agent templates in prompts.md (diverge, cross, curator, judge, developer, stress) wrap the brief content within === THE BRIEF === and === END OF THE BRIEF === delimiters to prevent instructions inside the brief from being mistaken for agent instructions.
Capability inventory: The orchestrator uses python3 for state management and the Agent tool to spawn sub-agents. Sub-agents (developer and stress phases) have access to web search tools for fact-checking.
Sanitization: The skill relies on structured delimiters rather than complex sanitization, which is appropriate for its intended use as a collaborative brainstorming tool.
[COMMAND_EXECUTION]: The skill uses a local Python script (council.py) to manage the state of the multi-round brainstorming session. This script uses standard library modules and performs legitimate file system operations within the .council/ directory to track progress and store sub-agent outputs.

council