The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided markdown files (Workflow A) which could potentially contain malicious instructions. The skill has access to powerful tools like Bash and Edit. · Ingestion points: Reads existing markdown files in Workflow A to generate HTML companions. · Boundary markers: The skill relies on a 'cardinal rule' to ensure output parity but does not specify technical delimiters to prevent the agent from obeying instructions embedded in source files. · Capability inventory: The skill uses Read, Write, Edit, Glob, Grep, and Bash tools across its workflows. · Sanitization: No explicit sanitization or filtering of input markdown content is described. · Mitigation: The instructions mandate that source files are read-only by default and requires explicit human confirmation before the agent performs any edits to the markdown source.
[EXTERNAL_DOWNLOADS]: The skill allows for the inclusion of CDN-hosted libraries (such as D3.js or Three.js) within the generated HTML companion files. These references target well-known and trusted technology services for the purpose of enabling interactive data visualizations.

interactive-doc