The Agent Skills Directory

[SAFE]: The skill consists of natural language instructions for a development workflow. No malicious code, obfuscation, or unauthorized data access patterns were detected.
[INDIRECT_PROMPT_INJECTION]: The workflow describes ingesting external 'evidence' (PoC code) and writing it to a repository. While this is an ingestion surface, the instructions mitigate risk by mandating that the agent identifies authorized scopes and verifies results against expected behaviors, maintaining human-in-the-loop approval as a prerequisite.

build-prototype