baz-codebase-exploration

Pass

Audited by Gen Agent Trust Hub on Aug 24, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill processes content from external repositories, which serves as an indirect prompt injection attack surface.
  • Ingestion points: Data enters the agent context through 'remote_grep', 'remote_file_search', and subsequent file-read/fetch operations.
  • Boundary markers: The instructions do not specify the use of delimiters or 'ignore' directives when processing the retrieved code content.
  • Capability inventory: The agent uses the findings to 'plan a feature', 'design a change', and 'scope work', which may involve creating code or planning cross-repo coordination.
  • Sanitization: There is no requirement for sanitizing or escaping the retrieved external content before it is used in the planning process.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 24, 2026, 01:51 PM
Security Audit — agent-trust-hub — baz-codebase-exploration