codebase-study-guide
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill systematically explores a codebase, including untrusted external data such as README files, architecture docs, and inline comments, and processes this information to generate a study guide. This creates a surface for indirect prompt injection where malicious instructions embedded in the codebase could influence the agent's behavior during the guide generation process.
- Ingestion points: Step 2 of the workflow explicitly directs the agent to read 'Existing docs — README, CLAUDE.md, architecture docs, ADRs, inline comments' and explore all file contents.
- Boundary markers: The skill does not provide delimiters or instructions to treat codebase content as untrusted, increasing the risk that the agent may follow instructions found within the code.
- Capability inventory: The skill uses file-read tools and the Explore agent, and has the capability to write a Markdown file ('.claude/docs/study-guide.md') to the local filesystem.
- Sanitization: No sanitization, escaping, or validation of the ingested content is performed before interpolation into the final document.
- [DATA_EXFILTRATION]: The skill instructs the agent to investigate system configuration, including 'env vars' and 'config files' (SKILL.md, Step 2). While intended for architectural analysis, this creates a risk of sensitive data exposure if the agent extracts credentials or secrets from these files and includes them in the generated study guide, which is then written to a permanent file.
Audit Metadata