The Agent Skills Directory

[SAFE]: The skill is a collection of prompt templates and documentation. It does not perform any file system operations, network calls, or command execution.
[PROMPT_INJECTION]: Analysis of the stance prompts in references/stances.md shows that the instructions are designed to reinforce safety. While the text uses phrases like 'override your stance,' it does so to ensure the agent refuses to support harmful or technically infeasible proposals, acting as a defensive measure rather than an attack vector.
[INDIRECT_PROMPT_INJECTION]: As a tool for analyzing external proposals, the skill possesses an ingestion surface. However, it mitigates risk by instructing the agent to maintain objectivity, verify claims against documentation/code, and adhere to mandatory safety guardrails even when performing biased roles (like the Advocate or Critic stances).

perspectives