The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection (Category 8) by interpolating untrusted data (user question) and semi-trusted local file content into prompts for sub-agents.
Ingestion points: The user-supplied <question> argument and the contents of SKILL.md and reference.md files from other skills found via the Glob tool in the ../../skills/*/ directory.
Boundary markers: User input and file paths are wrapped in double quotes within the Qualification Prompt and Council Statement Prompt templates in reference.md (e.g., "{question}").
Capability inventory: The skill uses the Agent tool to spawn sub-processes for philosopher personas and the Write tool to archive session results in the current working directory.
Sanitization: No explicit sanitization, filtering, or validation of the user question or the content of the discovered skill files is performed before they are processed by the sub-agents.

council