The Agent Skills Directory

[SAFE]: The skill provides a professional framework for architectural and security auditing. It focuses on identifying best practices and identifying risks like hardcoded credentials or lack of input validation in the target code.
[PROMPT_INJECTION]: The skill is subject to indirect prompt injection because it processes untrusted source code provided by the user.
Ingestion points: The agent reads all files within a specified directory or module (SKILL.md).
Boundary markers: Not present. There are no instructions to ignore natural language instructions found within code comments or string literals.
Capability inventory: The skill primarily uses file read operations and generates markdown reports.
Sanitization: None. The skill directly interprets the code it reads to produce its evaluation.
Context: While instructions in analyzed code could attempt to bias the scorecard, the skill's requirement for evidence-based citations (file and line numbers) mitigates the risk of the agent hallucinating or following hidden commands without a traceable reason.

software-evaluation