The Agent Skills Directory

[SAFE]: The skill is a purely informational resource providing architectural patterns for agentic evaluation. It does not perform any actions or include malicious instructions.
[NO_CODE]: No executable files or scripts are included. All logic is presented as non-functional examples within the SKILL.md file for educational purposes.
[PROMPT_INJECTION]: The reflection patterns described involve re-processing AI-generated data, which represents an indirect prompt injection surface.
Ingestion points: The 'output' and 'critique' variables in SKILL.md.
Boundary markers: Not included in the conceptual snippets.
Capability inventory: The 'run_tests' pattern implies code execution.
Sanitization: Not present in the provided templates.
[COMMAND_EXECUTION]: The skill suggests a pattern for executing generated code via a conceptual 'run_tests' function. While purely instructional here, this identifies an attack surface that would require secure sandboxing in a real-world implementation.

agentic-eval