generating-codebase-wiki
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes external and untrusted codebase files to generate documentation, creating a surface for potential indirect prompt injection.
- Ingestion points: As described in research-protocol.md, the agent reads source files, project manifests, and entry modules from a user-provided repository root.
- Boundary markers: The skill requires the use of page.template.html and specific citation tags like
<code class="cite">to delimit information derived from the codebase. - Capability inventory: The agent executes shell commands for file management (mkdir, cp) and runs a Node.js verification script (verify-cites.mjs). It is also capable of writing generated HTML to the local filesystem.
- Sanitization: The citation protocol in cite-write.md focuses on grounding claims in real code symbols for accuracy, but there is no explicit automated sanitization of codebase content before it is included in the documentation output.
- [COMMAND_EXECUTION]: The skill instructions include executing local scripts and shell commands to facilitate the documentation workflow.
- The init-site.sh shell script is used to scaffold the output directory and copy CSS/JS assets.
- The verify-cites.mjs Node.js script performs static analysis on the generated site to ensure citations correctly reference existing files and line ranges within the repository.
Audit Metadata