The Agent Skills Directory

[SAFE]: The skill is a legitimate reasoning utility. No malicious code, hardcoded credentials, or exfiltration attempts were detected.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input (problem statements and solution texts), creating a surface for indirect prompt injection. 1. Ingestion points: The skill reads the "original problem statement" and "provided solution" from files or user input using the Read tool. 2. Boundary markers: None; the skill uses descriptive Markdown headers but does not define unique random delimiters for the external content. 3. Capability inventory: The skill has read-only filesystem access via Read, Grep, and Glob. 4. Sanitization: None; the external text is processed directly as part of the reasoning analysis. This surface is assessed as safe because the available tools are read-only and lack network or write capabilities, preventing any significant impact from injected instructions.

reasoning-verifier