The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (a story "IDEA") and interpolates it into a project brief and subsequent code generation tasks.
Ingestion points: User-supplied IDEA string defined in SKILL.md.
Boundary markers: No explicit boundary markers or "ignore instructions" warnings are utilized during the interpolation of the idea into brief.md.
Capability inventory: The skill possesses extensive capabilities to write files (brief.md, index.tsx, shots.ts, cast.ts), execute shell commands via the dapi CLI, and invoke MCP tools for rendering.
Sanitization: The skill does not describe any specific sanitization or validation of the input idea before processing it.
[DYNAMIC_EXECUTION]: The core functionality of the skill involves the agent generating, modifying, and rendering TypeScript and JSX code based on a provided engine and templates.
Evidence: The workflow described in SKILL.md (Step 3: Blocking) involves replacing look tests with real shot tables and character rigs in .ts and .tsx files which are then executed by the rendering engine.
[COMMAND_EXECUTION]: The skill makes extensive use of the dapi CLI and MCP tools to interact with the local environment and the Diffusion Studio application.
Evidence: SKILL.md lists tools like open, check, capture, and logs. example/fish-heist/package.json contains scripts for media probe, grab, listen, whoami, and report using the dapi command.
[EXTERNAL_DOWNLOADS]: The installation instructions reference downloading a repository from a remote source.
Evidence: README.md instructs users to clone https://github.com/diffusionstudio/open-projects to install the skill.

take-on-me