improve-codebase-architecture
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from the codebase being analyzed, which is potentially untrusted, to generate reports and guide the design process.
- Ingestion points: Codebase contents (via sub-agent exploration),
CONTEXT.md, and architectural decision records indocs/adr/. - Boundary markers: No specific delimiters or safety instructions are defined to separate untrusted codebase data from the agent's instructions.
- Capability inventory: Writing to the filesystem, opening local files in a browser, and spawning multiple sub-agents using the
Agenttool. - Sanitization: The skill does not describe mechanisms to sanitize or escape data before rendering it in the HTML report, which explicitly initializes the Mermaid library with a 'loose' security setting.
- [EXTERNAL_DOWNLOADS]: The skill generates HTML reports that reference external resources from well-known services.
- Fetches styling configuration from Tailwind CSS's CDN.
- Downloads the Mermaid visualization library from the jsDelivr CDN repository.
Audit Metadata