The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill operates on existing course content and potentially external materials, creating a surface where ingested data could influence agent behavior.
Ingestion points: Content is fetched from course storage via read_stage and list_scenes. The instructions also note that the skill may use "attached materials or web access" to ground edits.
Boundary markers: The instructions focus on precise JSON Pointer addressing (e.g., /content/canvas/elements/N/content) to isolate changes, providing structural separation, though explicit prompt boundaries for content-level instructions are not specified.
Capability inventory: The skill possesses significant capabilities including database modification (patch_stage), deck restructuring (edit_deck), page regeneration (generate_scene), and audio synthesis (generate_tts).
Sanitization: Instructions require the agent to perform 'surgical' edits, match existing styles, and conduct verification passes using render_scene_preview and list_scenes to ensure integrity.

pro-editing