The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data from uploaded .pptx files through the import_pptx and read_stage tools.
Ingestion points: Content enters the agent's context from PowerPoint slides and speaker notes (referenced in SKILL.md Step 1 and Step 3).
Boundary markers: No explicit technical delimiters or boundary markers are defined to isolate the imported text from the agent's instructional context.
Capability inventory: The skill uses powerful tools to modify the environment, including patch_stage, edit_deck, and generate_actions (referenced in Step 3 and Step 4).
Sanitization: While the skill instructs the agent to treat speaker notes as 'drafts' and rewrite them rather than following them verbatim, this is a procedural instruction rather than a technical sanitization or filtering mechanism.
[DYNAMIC_EXECUTION]: The skill instructs the agent to dynamically load and follow a secondary set of instructions at runtime.
Evidence: Step 3c explicitly directs the agent to "Load pro-editing first" by reading its SKILL.md from a variable path and following its internal logic. This creates a dependency on the integrity and safety of the external pro-editing skill content.

pptx-import