agents-md-generator
Pass
Audited by Gen Agent Trust Hub on Mar 31, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill utilizes established, read-only command-line utilities for repository analysis, including tokei, ripgrep (rg), and tree. These tools are used strictly to inspect the directory structure and measure code volume for documentation purposes.- [SAFE]: Explicit security boundaries are defined in the read_only_commands.md file, which instructs the agent to never read dependency lock files (e.g., package-lock.json, pnpm-lock.yaml) to prevent the accidental exposure of transient dependency data or environment-specific configurations.- [PROMPT_INJECTION]: Indirect Prompt Injection Surface:
- Ingestion points: Repository source code, configuration manifests (e.g., package.json, pyproject.toml), and documentation files (README.md) read during the analysis phase.
- Boundary markers: Absent; the agent processes file content directly through standard search and read tools.
- Capability inventory: Repository structure visualization (tree, find), content searching (rg, grep), and paginated file reading (sed).
- Sanitization: Absent; the skill relies on the agent's internal reasoning to identify patterns and conventions from raw file content. This risk is inherent to the skill's primary function of codebase analysis and is mitigated by behavioral agreements specifying minimal changes and adherence to existing public APIs.
Audit Metadata