The Agent Skills Directory

[SAFE]: The skill is designed for qualitative analysis and reporting. It operates entirely within the provided project context, reading local files such as token definitions, component source code, and documentation to compare them against established industry benchmarks.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a data ingestion surface in SKILL.md (Step 0) where untrusted content from local project files (component source code, documentation sites, token files, and release history) is read into the agent context. While no explicit boundary markers or sanitization are defined, the skill's capabilities are limited to generating a markdown report for user review, which significantly mitigates the risk of embedded instructions influencing external systems or critical operations.

system-benchmark