The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external idea documents and wireframe HTML files to generate presentation content. This creates a surface for indirect prompt injection where malicious input could try to manipulate the AI's output or rendering logic.
Ingestion points: The skill reads user-provided project documents and wireframe HTML files from the workspace.
Boundary markers: Output is generated via a structured JSON specification and text is escaped during the HTML rendering phase.
Capability inventory: The skill uses Python scripts to perform rendering and automated validation using a headless browser.
Sanitization: The generation logic includes strict checks to ban script tags in wireframes and prohibits dangerous JavaScript features like eval, innerHTML, or network requests (fetch, XMLHttpRequest).
[EXTERNAL_DOWNLOADS]: The skill automates the setup of a validation environment by installing the playwright library and the Chromium browser binary from well-known sources.
Evidence: SKILL.md contains instructions to pip install playwright and playwright install chromium inside a local virtual environment.
[COMMAND_EXECUTION]: The skill utilizes the agent's Bash tool to perform setup tasks (creating a venv, installing dependencies) and to run the generation and validation scripts.

create-pitch-deck