The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill accepts user input via $ARGUMENTS and interpolates it into the instructions for multiple sub-agent personas (e.g., Ada, Aristotle, Socrates) during the council workflow. This lack of sanitization or robust boundary markers allows for potential instruction overrides within the sub-agent contexts.\n- Ingestion points: The workflow in SKILL.md (Workflow Step 2: Round 1 and Step 3: Round 2) reads $ARGUMENTS and passes it to the sub-agent prompts.\n- Boundary markers: The prompt templates in SKILL.md (e.g., the round 1 packet) use simple label-based interpolation (Problem: {problem}) without explicit delimiters or specific instructions to the sub-agents to ignore embedded commands.\n- Capability inventory: The skill utilizes powerful agent orchestration features (e.g., spawn_agent, oracle, or parallel contexts) to delegate deliberation tasks to independent sub-agents.\n- Sanitization: No explicit sanitization, filtering, or escaping of the $ARGUMENTS input is performed before it is presented to the sub-agents.

council