build-with-cracked

Pass

Audited by Gen Agent Trust Hub on Sep 2, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill operates by ingesting and processing data from over 60,000 external tools, including social media platforms and web scrapers, which creates a significant surface for indirect prompt injection if the retrieved content contains malicious instructions.\n
  • Ingestion points: The agent receives external data through the results of the POST /v1/run endpoint.\n
  • Boundary markers: There are no explicit delimiters or specific instructions for the agent to disregard instructions embedded within the retrieved tool outputs.\n
  • Capability inventory: The skill includes functionality for network interaction via various tools and file management through a dedicated workspace (/v1/files).\n
  • Sanitization: The documentation does not specify sanitization or escaping procedures for external content prior to agent processing.\n- [EXTERNAL_DOWNLOADS]: The skill documentation refers to several vendor-specific packages and resources required for operation.\n
  • Python packages: Includes cracked-ai, cracked-ai-langchain, cracked-ai-crewai, and cracked-ai-openai-agents.\n
  • Node.js packages: Includes cracked-ai-sdk and cracked-ai-vercel.\n
  • External references: Links to official documentation, schemas, and live skill definitions hosted on cracked.ai.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 2, 2026, 07:42 PM
Security Audit — agent-trust-hub — build-with-cracked