The Agent Skills Directory

[SAFE]: The skill is designed as a conversational framework for decision-making. It contains explicit instructions that prevent the agent from implementing changes, migrating data, or writing files, stating that 'confirmation of understanding is not authorization to implement or publish.' All operations are limited to text-based interaction and local file reading for context.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface by reading environment facts (such as existing project plans or CI workflows) to formulate its interview questions. Although this involves processing untrusted local content, the risk is negligible because the skill lacks executable capabilities (no shell access, no network exfiltration, no file-write tools) and uses a structured round-based questioning format that keeps the user in control of all final decisions.

refine