guidelines-advisor

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external smart contract codebases, which represents a surface for potential indirect prompt injection if instructions are hidden in the analyzed files.
  • Ingestion points: The agent reads contract source files, module code, and project documentation during the discovery phase (SKILL.md).
  • Boundary markers: The assessment process is structured into five distinct phases (Discovery, Documentation, Architecture, Implementation, Recommendations) to maintain focus and context boundaries.
  • Capability inventory: The skill instructions recommend utilizing static analysis tools like Slither and standard file system search utilities to examine code implementation.
  • Sanitization: The methodology requires findings to be presented with specific file and line references for verification, and includes a rationalization step to prevent skip errors or misinterpretations.
  • [COMMAND_EXECUTION]: The instructions suggest using established code analysis tools and search utilities to identify vulnerabilities and map architectural patterns.
  • Evidence: The skill references using Slither printers for generating diagrams and using codebase search for identifying common pitfalls (ASSESSMENT_AREAS.md). These tools are standard for the security auditing domain and the vendor context.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 09:17 PM
Security Audit — agent-trust-hub — guidelines-advisor