The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted repository code to generate architecture diagrams and a navigable HTML deliverable. This creates a surface where malicious identifiers or comments in the analyzed codebase could be interpolated into the generated output.
Ingestion points: Source code files and repository structures read during the analysis phase (SKILL.md).
Boundary markers: No explicit delimiters are specified to isolate untrusted code content within the generated documentation.
Capability inventory: The skill writes generated content to the file system, including a full HTML deliverable (SKILL.md).
Sanitization: The instructions do not specify HTML escaping or sanitization of extracted code identifiers before they are inserted into the architecture.html template.
[EXTERNAL_DOWNLOADS]: The generated documentation deliverable fetches the Mermaid.js library from a well-known CDN for runtime rendering.
Source: The assets/template.html file includes a script tag targeting https://cdn.jsdelivr.net/npm/mermaid@11/dist/mermaid.min.js to render the architecture diagrams.

blueprint