impeccable

Warn

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: MEDIUMINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests design and product context from PRODUCT.md and DESIGN.md files in the project root. This creates a surface where instructions embedded in project files can influence agent behavior.
  • Ingestion points: Project configuration data is read by scripts/load-context.mjs.
  • Boundary markers: Ingested markdown content is consumed without explicit delimiters or warnings to ignore embedded commands.
  • Capability inventory: The skill has extensive file-writing capabilities and can execute local Node.js utility scripts.
  • Sanitization: No sanitization or schema validation is performed on the ingested markdown strings.
  • [DYNAMIC_EXECUTION]: To enable its Live Mode iteration features, the skill performs automated code injection. The scripts/live-inject.mjs script modifies project HTML, JSX, and TSX files to insert a script tag connecting to a local helper server, and it relaxes Content-Security-Policy (CSP) headers to allow localhost connections. Additionally, scripts/live-browser.js compiles and runs WebGL fragment shaders in the user's browser.
  • [COMMAND_EXECUTION]: The skill automates its workflow by spawning background processes (scripts/live-server.mjs) and executing multiple local Node.js scripts for maintenance, project analysis, and task orchestration. It also executes git commands to verify repository status.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 15, 2026, 05:03 AM
Security Audit — agent-trust-hub — impeccable