frame
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and structure untrusted data from external Figma files, including component descriptions, property values, and variable definitions, which could be manipulated to include malicious instructions.\n
- Ingestion points: Data is brought into the agent context via
get_design_context,get_variable_defs,get_metadata, andget_figjamtools.\n - Boundary markers: The skill defines strict handoff formats (Markdown and YAML blocks) to separate design data from instructional context, though these do not prevent adversarial injection.\n
- Capability inventory: The skill has significant capabilities, including writing to the Figma canvas (
use_figma), creating new files (create_new_file), and syncing Code Connect mappings.\n - Sanitization: The reference documentation (
reference/figma-mcp-server-ga.md) contains a quality guardrail instructing the agent to treat all tool outputs and component descriptions as untrusted input.\n- [DYNAMIC_EXECUTION]: The skill facilitates the generation and execution of scripts targeting the Figma Plugin API to modify design canvases.\n - Evidence: The
use_figmatool allows the agent to execute code snippets (e.g., usingfigma.*Async()calls) to create or mutate layers, variables, and components.\n - Context: This behavior is central to the skill's purpose of enabling 'code to canvas' workflows and is mitigated by the requirement for explicit user confirmation before any write operation is performed.\n- [EXTERNAL_DOWNLOADS]: The skill connects to and fetches resources from official Figma infrastructure.\n
- Evidence: Instructions establish a connection to the official remote MCP endpoint at
https://mcp.figma.com/mcpand reference various developer documentation pages ondevelopers.figma.com.\n - Context: These interactions target a well-known service and are documented as standard operational requirements.
Audit Metadata