designlang-design-extract

Pass

Audited by Gen Agent Trust Hub on Apr 16, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill documentation relies on executing the 'designlang' CLI tool via 'npx' and suggests global installation via 'npm install -g'. These operations execute third-party code from the npm registry with access to the user's environment.\n- [EXTERNAL_DOWNLOADS]: The skill triggers the download of the 'designlang' package from the npm registry and references a third-party GitHub repository ('Manavarya09/design-extract') for skill installation.\n- [CREDENTIALS_UNSAFE]: The documentation provides explicit examples for passing sensitive data, such as session cookies, API keys, and bearer tokens, directly as command-line arguments using the '--cookie' and '--header' flags. This can lead to credential exposure in process lists and shell history.\n- [PROMPT_INJECTION]: The skill facilitates indirect prompt injection by design.\n
  • Ingestion points: Data is crawled from external websites and saved into 'AI-optimized markdown' files (e.g., '*-design-language.md') for agent consumption.\n
  • Boundary markers: Absent; there are no instructions or delimiters to isolate the untrusted external content from the agent's instructions.\n
  • Capability inventory: The skill has remote code execution ('npx') and extensive file-system write capabilities.\n
  • Sanitization: There is no evidence of sanitization to remove malicious instructions embedded in crawled content before it is presented to the LLM.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 16, 2026, 03:14 AM