The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill manages dependencies on the GPUI framework by referencing the official Zed Industries GitHub repository. These dependencies are pinned to specific Git revisions, ensuring build stability and preventing the accidental introduction of unverified code from upstream updates.
[COMMAND_EXECUTION]: Includes shell scripts such as inspect_gpui_project.sh and validate_reference_app.sh that utilize standard developer tools including cargo, git, and ripgrep. These scripts are used locally to analyze project structure and verify implementation correctness, which is standard practice for software development skills.
[DATA_EXPOSURE]: Provides a workflow for design-to-code translation using an MCP integration with the Paper.design desktop app. This allows the agent to retrieve design specifications, tokens, and assets from the user's local environment to facilitate UI construction, following a least-privilege model via a local endpoint.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external design data (JSX, text content, metadata) via the Paper MCP tool. While this presents an ingestion surface for untrusted content, the instructions guide the agent to treat this data as structural design evidence rather than executable instructions.
Ingestion points: Design nodes and text retrieved from Paper.design files via the paper MCP tool.
Boundary markers: The skill instructions emphasize treating the data as visual and structural facts but do not include explicit 'ignore embedded instructions' markers for all input fields.
Capability inventory: Access to local development tools (cargo, rustc) and filesystem inspection (grep, rg).
Sanitization: The skill relies on the agent's internal safety filters and the specific mapping of design tokens to handle external strings.

build-gpui-apps