codebase-analysis

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The codebase-analysis skill functions by scanning external and potentially untrusted repository files, such as source code comments and requirement sidecars, and includes them directly in downstream artifacts. This creates an open ingestion surface for malicious data or instructions embedded within codebases.
  • Ingestion points: The skill reads untrusted text from codebase files (source comments, headers, *.trace, requirements.csv, and swr_map.json) as detailed in section 3a of SKILL.md.
  • Boundary markers: No explicit delimiters, data isolation mechanisms, or protective instructions are specified to prevent the model from interpreting or being misdirected by text found in parsed comments.
  • Capability inventory: The skill possesses file system write capabilities, writing structured map files to .autonomousguy/CODEBASE_MAP.md and generating interactive HTML reports inside the project's analysis/ directory.
  • Sanitization: There are no requirements or instructions to sanitize, strip HTML tags, or escape the parsed strings before interpolating them into the markdown files or the references/html-report-template.html template.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 10:20 AM
Security Audit — agent-trust-hub — codebase-analysis