The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the execute_code tool to run JavaScript snippets that interface with the Penpot plugin API. This is the intended functional mechanism for design automation, including board creation, layout configuration, and component instantiation.
[PERSISTENCE]: Project-specific state and progress are managed through the setSharedPluginData API to maintain a run ledger. This mechanism is used appropriately to track the incremental build process across multiple turns and does not exhibit malicious persistence patterns.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided design briefs which serve as an untrusted input surface. This is mitigated through a structured workflow requiring user approval of the planning phase and visual confirmation of each generated design section.
Ingestion points: Design briefs are ingested during Phase 0 analysis.
Boundary markers: The agent is instructed to confirm the interpreted brief, style profile, and section list with the user before performing any design mutations.
Capability inventory: The skill has capabilities to create and modify design shapes, apply tokens, and fetch remote assets into the Penpot canvas via uploadMediaUrl.
Sanitization: The skill relies on iterative human review and visual self-critique (Phase N+1) to validate the agent's output against the user's intent.

penpot-build-screen