The Agent Skills Directory

[PROMPT_INJECTION] (LOW): The skill's session management logic includes a pattern for 'summoning' sub-agents that is vulnerable to indirect prompt injection.
Ingestion points: In SESSIONS.md, the template for specialist agents directly interpolates the [Insert full question and context] string.
Boundary markers: Absent. The template does not employ delimiters or specific instructions to treat the interpolated content as untrusted data.
Capability inventory: Specialist agents are granted significant capabilities, including network access via WebSearch and local file-writing to the .council/ directory.
Sanitization: Absent. The instructions do not specify any validation or escaping for the user-provided question or context before it enters the agent's prompt.
[COMMAND_EXECUTION] (SAFE): The skill utilizes basic shell commands for session state management.
Evidence: SESSIONS.md contains commands such as mkdir and cat to initialize the .council directory structure.
Analysis: These commands are used for legitimate organization purposes, are restricted to local paths, and do not involve executing untrusted remote code or escalating privileges.

the-council