The Agent Skills Directory

[SAFE]: The skill serves as a documentation and pattern library for LLM evaluation. It contains no malicious code, obfuscation, or unauthorized data access patterns.\n- [INDIRECT_PROMPT_INJECTION]: The provided code examples for 'LLM-as-Judge' (e.g., llm_judge_quality) demonstrate patterns where untrusted data is interpolated into prompts. While this represents a potential surface for indirect prompt injection, it is consistent with the skill's educational purpose and does not constitute a vulnerability in the skill itself.\n
Ingestion points: references/details.md defines functions like llm_judge_quality that take response and question as arguments.\n
Boundary markers: Absent in the example templates.\n
Capability inventory: Examples use standard API clients like anthropic. No dangerous local capabilities are exposed.\n
Sanitization: Not included in the illustrative examples.

llm-evaluation