kta-compose-design-tokens
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill architecture ingests untrusted design specifications, creating a potential surface for indirect prompt injection attacks.
- Ingestion points: Figma URLs, Stitch project IDs, and Claude-generated design specs (HTML, SVG, or Markdown artifacts) parsed via
references/claude-design-integration.mdandreferences/figma-mcp-integration.md. - Boundary markers: The skill implements a mandatory 'Input Gate' in
SKILL.mdto scan for specific markers and validate seed data before proceeding. It also contains instructions to explicitly refuse out-of-scope requests. - Capability inventory: The skill manages file system writes (Kotlin source code) and delegates complex UI construction tasks to an implementation agent (
kta-compose-developer). - Sanitization:
scripts/generate-kotlin-tokens.pyincludes logic to sanitize identifiers and handle Kotlin reserved words to prevent code injection during generation. - [EXTERNAL_DOWNLOADS]: The documentation describes and encourages the use of external design services and development tools.
- Details:
references/figma-mcp-integration.mdandreferences/stitch-mcp-integration.mdprovide instructions for integrating with official and community MCP servers vianpxand specific remote URLs (e.g.,mcp.figma.com). These references are neutrally documented as part of the tool's intended setup workflow for well-known services.
Audit Metadata