sketch-design-to-code

Pass

Audited by Gen Agent Trust Hub on Jun 26, 2026

Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local shell commands and development scripts, including npm run dev, yarn ios, and mkdir -p (mentioned in SKILL.md and references/validation.md), as part of its workflow to set up, run, and validate the generated code.\n- [REMOTE_CODE_EXECUTION]: The skill utilizes the run_code tool from the Sketch MCP to execute JavaScript snippets within the Sketch application environment (as detailed in SKILL.md and references/extraction.md) for deep inspection of design properties.\n- [DATA_EXFILTRATION]: The skill reads sensitive project configuration and manifest files (e.g., package.json, build.gradle, Podfile, and other config files in SKILL.md Step 2) to align the generated code with existing project conventions.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the following evidence chain:\n
  • Ingestion points: Untrusted data enters the context via get_layer_tree_summary (text snippets, names) and get_symbol_overrides in SKILL.md and references/extraction.md.\n
  • Boundary markers: Absent; there are no instructions to use delimiters or ignore embedded instructions in the ingested data.\n
  • Capability inventory: The skill has significant capabilities including shell command execution (npm run dev) and JavaScript execution via run_code in SKILL.md and references/validation.md.\n
  • Sanitization: Absent; the instructions do not specify any validation or filtering of external content from the Sketch document.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 26, 2026, 06:45 AM
Security Audit — agent-trust-hub — sketch-design-to-code