The Agent Skills Directory

[SAFE]: The skill defines a cognitive workflow for clarifying ambiguous tasks. It focuses on identifying 'knowns' and 'unknowns' to build a shared understanding between the user and the agent. The instructions do not involve sensitive operations, network access to unknown domains, or execution of external scripts. All referenced files are local templates and configuration files provided within the skill package.- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-provided task descriptions which naturally serve as an ingestion point for untrusted data. However, the workflow enforces a structured diagnosis and confirmation checkpoint ('Core Contract') that encourages critical evaluation of the task parameters rather than immediate execution. The skill lacks dangerous capabilities such as arbitrary file writes or remote code execution, limiting the impact of any potential injection.

unknowns-first