The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data in the form of plans, decisions, or ideas which could potentially contain instructions intended to influence the agent's behavior during the 'grilling' process.
Ingestion points: The skill ingests user input via the interview process described in SKILL.md (e.g., plans, decisions, or ideas provided by the user).
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore potentially malicious content embedded within the user's plan.
Capability inventory: The skill explicitly instructs the agent to use tools and sub-agents to query the environment (filesystem, etc.) to verify facts related to the user's plan as seen in SKILL.md.
Sanitization: No sanitization or validation of the user-provided data is specified before it is processed or used to trigger sub-agent tasks.
[SAFE]: The skill instructions are purely behavioral and intended for productivity. The installation command provided in the README.md uses a known agent platform CLI (skills). The logic for dispatching sub-agents for fact-finding is a standard capability of advanced AI agents and does not constitute a security violation in this context.

grilling