The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize data from various project files and research notes, creating an attack surface for instructions embedded in that data.
Ingestion points: Processes user-specified files, conversation history, and the results of /research and /wayfinder tool invocations.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' commands when processing external material.
Capability inventory: The skill possesses file-write capabilities in the .scratch/ directory, executes shell commands via uvx, and can invoke multiple other system tools (/research, /wayfinder, /pr-lens, /diagram, /browse).
Sanitization: There is no explicit sanitization or escaping of the ingested data before it is incorporated into prompts or generated scripts.
[COMMAND_EXECUTION]: The skill executes external command-line tools as part of its primary workflow.
Evidence: Uses uvx to run the Manim animation engine for video rendering and suggests the use of the Homebrew package manager (brew) for installing system dependencies.
[DYNAMIC_EXECUTION]: The skill generates and subsequently executes executable code at runtime.
Evidence: It dynamically creates a video/scene.py Python script based on user-provided content and then executes it using Manim to produce video files.
[EXTERNAL_DOWNLOADS]: The generated HTML output utilizes external libraries from established content delivery networks.
Evidence: References Tailwind CSS, Scrollama, and Mermaid.js from tailwindcss.com, unpkg.com, and jsdelivr.net respectively to provide styling and diagrams in the generated explainers.

writing-for-humans