The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs shell commands to detect local configuration files (EXTEND.md) and verify the existence of output directories. Additionally, it utilizes the npx -y bun command to execute internal utility scripts (merge-to-pptx.ts and merge-to-pdf.ts) which are responsible for final document assembly.
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to ingest and analyze arbitrary user-provided text to generate presentation outlines and image generation prompts, it possesses an inherent attack surface for indirect prompt injection. Malicious instructions hidden within the source material could potentially influence the LLM during the prompt engineering phase.
Ingestion points: User-provided content stored in source.md or pasted into the session.
Boundary markers: The instructions lack explicit delimitation or strict 'ignore external instruction' framing for the content processing step.
Capability inventory: The skill has access to file system read/write operations, shell command execution via Bun, and image generation capabilities.
Sanitization: The skill does not explicitly describe input sanitization or validation logic for the content analysis stage.
[EXTERNAL_DOWNLOADS]: The workflow relies on npx -y bun, which may download the Bun runtime if it is missing from the environment. Both Bun and the utilized libraries (pdf-lib, pptxgenjs) are well-known technology services and standard development dependencies.

canghe-slide-deck