sync
Pass
Audited by Gen Agent Trust Hub on Nov 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to verify project integrity and perform visual rendering. Specifically, it directs the agent to 'run the project's checks (lint, typecheck, tests, visual tests)' in
SKILL.mdandreferences/export-diff.md. Additionally,references/adapters/html.mdprovides a specific command for taking screenshots using a headless browser:"<chrome>" --headless=new --screenshot=guide-1280.png --window-size=1280,900 file://$PWD/guide/index.html.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from the project codebase which could contain malicious instructions designed to influence the agent's behavior.\n - Ingestion points: Files described in
references/detect.md, including CSS custom properties (tokens.css), Tailwind configurations (tailwind.config.*), JSON token files, theme objects in JS/TS/Rust/Swift, and component stories (*.stories.tsx).\n - Boundary markers: The skill implements multi-stage workflows ('Inventory, then stop' in
SKILL.mdandreferences/inventory.md, and 'Propose, then stop' inSKILL.mdandreferences/export-diff.md) intended to provide human oversight.\n - Capability inventory: File system modification (writing
design-sync.json,DESIGN.md, and editing source files), shell command execution (running tests and headless Chrome), and external tool interaction (Figma/Paper MCP tools).\n - Sanitization: There is no explicit content sanitization or escaping described for the ingested code or token values beyond the human review steps.- [DYNAMIC_EXECUTION]: The skill involves the dynamic creation and execution of content. The
references/adapters/html.mdfile describes generating a local documentation site (guide/index.html,guide/foundations.html) using code-derived values and then rendering those pages using a shell command to produce visual screenshots.
Audit Metadata