The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates and executes local scripts to perform core processing tasks. It defines a Python script for extracting content from PowerPoint presentations and a Node.js script using Playwright to take snapshots of the rendered React application.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from external PowerPoint (.pptx) files to convert them into React code. This process relies on the agent to interpret untrusted text and map it to code primitives, which is a potential vector for indirect prompt injection attacks.
Ingestion points: Text, titles, and speaker notes extracted from user-provided .pptx files via the ppt conversion mode.
Boundary markers: The instructions do not specify any delimiters or safety instructions to separate untrusted slide content from the agent's instructions during the conversion process.
Capability inventory: The skill can write TypeScript files, execute shell commands, run Python/Node.js scripts, and deploy assets to Vercel.
Sanitization: The skill does not implement specific sanitization or escaping for the content extracted from external files before it is used in code generation.
[COMMAND_EXECUTION]: The skill runs shell commands to manage its development environment and file system. This includes using lsof to detect running dev servers, creating directories for exports and archives, and using environment variables for path resolution.
[EXTERNAL_DOWNLOADS]: The skill references several external packages and tools, including python-pptx, playwright, imagemagick, img2pdf, and the Vercel CLI. These well-known resources are used for file processing and deployment.
[DATA_EXFILTRATION]: The export functionality includes a mode to deploy snapshots to Vercel, a well-known cloud service, which involves transmitting locally generated image assets and an HTML wrapper to an external cloud platform.

slide-deck