The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret data from _base.json (exported from a Figma plugin) and optionalContext (provided by a user). This creates a surface for indirect prompt injection where malicious instructions embedded in the design metadata or user context could influence the agent's interpretation or execution logic.
Ingestion points: SKILL.md specifies reading external content from {cachePath}/{componentSlug}-_base.json and consuming optionalContext as authoritative guidance.
Boundary markers: The instructions do not define strict boundary markers or delimiting protocols to separate user data from the agent's core instructions.
Capability inventory: The skill has the ability to write JSON files to the local disk and execute scripts via the figma_execute and use_figma tools.
Sanitization: No explicit sanitization, filtering, or validation rules for the ingested external data are described.
[DYNAMIC_EXECUTION]: The skill uses an 'escape hatch' (Step 3-delta) that allows the agent to generate and execute small JavaScript snippets at runtime to query the Figma API. While limited to 50 lines and read-only operations, this pattern involves dynamic code generation based on current task state.
[COMMAND_EXECUTION]: The skill utilizes external tools (figma_execute, use_figma) to run arbitrary JavaScript code within the Figma context for measurement purposes.

extract-structure