The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted design data (layers, text, and annotations) from Figma files, which could potentially contain malicious instructions intended to influence the agent's behavior during code generation.
Ingestion points: Figma URLs, file keys, node IDs, and design content such as text, hidden nodes, and prototype notes (SKILL.md, references/design-contracts.md).
Boundary markers: The "Trust Boundary" section in the primary skill file explicitly instructs the agent to treat design data as untrusted and prevents it from overriding the user's request or repository instructions (SKILL.md).
Capability inventory: The skill can write to the local file system (project components and configuration), execute CLI tools for project setup, and perform visual parity audits (SKILL.md, references/readiness.md).
Sanitization: The skill employs a design normalization process that converts raw design data into programmatic measurements and semantic structures, focusing on layout and content evidence rather than direct interpretation (references/design-normalization.md).
[EXTERNAL_DOWNLOADS]: The skill initiates the download and execution of the @primeuicom/cli package via npx for environment initialization and project setup (references/readiness.md).
[COMMAND_EXECUTION]: Shell commands are used for project readiness checks and initial configuration, specifically utilizing npx to run the Prime UI command-line interface for project setup (references/readiness.md).

figma-to-prime