The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data, such as codebase snippets and issue tracker content, to inform its decision-making subagents.
Ingestion points: The workflow involves collecting relevant files, snippets, and metrics from the working environment to provide context for the council.
Boundary markers: The subagent prompt template uses structured headers such as 'Question:' and 'Context:' to isolate untrusted data from the instructions.
Capability inventory: The skill utilizes session memory and standard persistence tools like knowledge-ops for its findings; it does not directly invoke shell commands or network operations.
Sanitization: No explicit sanitization of the gathered text is performed, relying instead on structural delimiters within the prompt.
[SAFE]: The instructions are focused on logical reasoning, perspective-taking, and structured dissent. The skill demonstrates good security posture by discouraging the use of shadow file paths for persistence and implementing anti-anchoring mechanisms to ensure subagent independence.

council