The Agent Skills Directory

[PROMPT_INJECTION]: The skill documents an architectural surface for indirect prompt injection via its codebase extraction functionality, which involves processing external source code to generate diagram labels and evidence nodes. 1. Ingestion points: Codebase analysis mentioned in SKILL.md and references/json-schema.md. 2. Boundary markers: Absent from the generated JSON instructions. 3. Capability inventory: The skill is restricted to file-writing operations for diagram assets. 4. Sanitization: Absent, as validation logic in references/validation.md is focused on structural schema integrity. This surface is considered safe given the skill's specific purpose of diagram generation.
[EXTERNAL_DOWNLOADS]: The integration guide in references/integration.md mentions external resources such as the @pma/viewer package hosted on the jsDelivr CDN and a server-side SVG rendering service. These are documented neutrally as standard integration paths for developers, and the documentation includes appropriate security context regarding SSRF and CORS for implementers.

pma-draw