The Agent Skills Directory

[SAFE]: The skill uses the term 'exploit' in its frontmatter description to refer to the 'explore/exploit' optimization paradigm, which is explicitly clarified in the instructions as the process of mining settled material. This is contextually appropriate and not a metadata poisoning attempt.
[SAFE]: The skill processes user-supplied markdown files ('raw material'), which represents a surface for indirect prompt injection. The skill manages this surface by using a strict, multi-step loop and limiting its capabilities to standard file reading and writing. Ingestion point: User-provided markdown file. Boundary markers: Absent. Capability inventory: File system read and write. Sanitization: Absent.
[SAFE]: The skill configuration includes a 'disable-model-invocation' flag and a policy prohibiting implicit invocation, which reduces the potential for complex tool-chaining attacks or unauthorized model usage. No network operations or unauthorized command execution patterns were detected.

writing-beats