The Agent Skills Directory

Indirect Prompt Injection (MEDIUM): The skill ingests untrusted content from markdown files to generate outlines and prompts for image generation. It lacks explicit sanitization or boundary markers to prevent embedded instructions in the source text from overriding the agent's behavior during the outline or prompt generation phases.
Ingestion points: SKILL.md Workflow Step 1 and 2 read external content from source.md or user-pasted text.
Boundary markers: None present in the outline templates or prompt generation logic.
Capability inventory: File system access, local script execution via bun, and image generation tool calls.
Sanitization: No sanitization or escaping of user-provided content is performed before interpolation into prompts.
Command Execution (LOW): The skill uses the agent to execute local TypeScript scripts via npx and bun. While these scripts are part of the skill's own package, the capability to execute commands based on generated file paths (slugs) poses a minor risk if path validation is not strictly enforced by the agent. (Evidence: SKILL.md Workflow Step 6; scripts/merge-to-pptx.ts, scripts/merge-to-pdf.ts).
External Downloads (LOW): The scripts rely on external Node.js libraries (pdf-lib and pptxgenjs) which may be downloaded or updated at runtime when using npx. These are standard libraries for these tasks but represent unverifiable dependencies at the time of execution. (Evidence: scripts/merge-to-pdf.ts, scripts/merge-to-pptx.ts).

baoyu-slide-deck