The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill is designed to ingest and analyze untrusted codebase data (such as CSS variables, JSON tokens, and Tailwind configurations) to build a design system plan.
Ingestion points: Phase 0 (Discovery) analyzes external files like *.tokens.json and variables.css (referenced in references/discovery-phase.md).
Capability inventory: The skill generates and executes JavaScript via the use_figma tool to create, modify, or delete nodes and variables in the Figma environment.
Sanitization & Boundary Markers: The workflow enforces a mandatory user checkpoint immediately after the discovery phase and before any write operations occur, ensuring that the human user reviews and approves the extracted data and the subsequent build plan.
[Dynamic Code Assembly]: The skill's primary operation involves the agent dynamically assembling and executing JavaScript code within the Figma Plugin API environment.
Evidence: SKILL.md and the references/ folder provide extensive script templates for the agent to adapt and run using the use_figma tool.
Context: This is the intended primary purpose of the skill, facilitating complex design automation. The risks associated with dynamic code generation are managed through a phased execution model and sequential validation steps.
[Session State Persistence]: The skill utilizes a state ledger stored in /tmp/dsb-state-{RUN_ID}.json to maintain progress across multiple turns and potential context resets.
Context: This is a standard and benign practice for long-running workflows, ensuring the agent can resume complex tasks without loss of progress or structural corruption.

figma-generate-library