searching-codebases

Pass

Audited by Gen Agent Trust Hub on May 7, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches repository source code as tarball archives from GitHub's official download service (codeload.github.com) based on user-provided URLs.
  • [EXTERNAL_DOWNLOADS]: Installs well-known, established libraries (scikit-learn, numpy, tree-sitter-language-pack) from the Python Package Index (PyPI) at runtime using the uv package manager.
  • [COMMAND_EXECUTION]: Utilizes ripgrep (rg) via subprocess calls to perform fast regex verification of search results. Commands are constructed as lists without shell interpolation, mitigating command injection risks.
  • [COMMAND_EXECUTION]: Executes the uv tool to manage local virtual environments and ensure required dependencies are available for search and mapping tasks.
  • [COMMAND_EXECUTION]: Interacts with a sibling skill (mapping-codebases) by executing its codemap.py script to generate structural context for semantic search enrichment.
  • [PROMPT_INJECTION]: While the skill ingests untrusted code from external repositories (Indirect Prompt Injection surface), it does so within a defined search scope and presents findings to the agent as code snippets. The risk is inherent to the tool's primary purpose and is managed by standard agent context boundaries.
Audit Metadata
Risk Level
SAFE
Analyzed
May 7, 2026, 03:55 AM