accessibility-annotations

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill contains logic patterns for runtime JavaScript execution within a Figma environment. As detailed in figma-tooling.md, it defines a 'recipe' for cloning nodes, dynamically updating properties using setProperties, and implementing font fallback handlers via figma.loadFontAsync. The agent is expected to generate and execute these scripts using an environment-provided tool like figma-use to interact with the Figma API.\n- [INDIRECT_PROMPT_INJECTION]: The skill operates by reading and processing content from external, untrusted sources which could contain malicious instructions designed to influence agent behavior during the annotation process.\n
  • Ingestion points: The project handoff document (referenced in handoff-template.md) and metadata/text from Figma canvas nodes (referenced in SKILL.md).\n
  • Boundary markers: Absent; there are no instructions to utilize delimiters or system prompts to ignore embedded instructions within the ingested handoff data.\n
  • Capability inventory: The skill possesses significant 'write' capabilities in the Figma document, including creating new nodes, modifying text properties, and altering layer hierarchies as specified in SKILL.md and figma-tooling.md via the Figma write tool.\n
  • Sanitization: Absent; the logic in figma-tooling.md and SKILL.md describes direct interpolation of data into component properties without mention of validation or output encoding.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 09:34 AM
Security Audit — agent-trust-hub — accessibility-annotations