Web3 Smart Contract Auditor
Fail
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: HIGHCREDENTIALS_UNSAFEREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: A hardcoded OpenRouter API key ('sk-or-v1-5d8cd33b...') was found in the .env file. This exposure allows unauthorized use of the associated account and services.\n- [REMOTE_CODE_EXECUTION]: The skill uses pickle.load() to process local data files in 'rag/index.py', 'rag/retriever.py', and 'rag/vectorstores/faiss_store.py'. This is a critical vulnerability that allows arbitrary code execution if an attacker can manipulate the files in the '.rag_store' directory.\n- [COMMAND_EXECUTION]: The rag_ingest tool, through 'rag/loader.py', can read arbitrary local files and remote URLs. This capability creates a significant risk of Server-Side Request Forgery (SSRF) and unauthorized access to sensitive system information.\n- [PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection because it processes untrusted external content and includes it in LLM prompts. (1) Ingestion points: 'rag/loader.py' (files and URLs). (2) Boundary markers: Weak markers like '[doc_id_chunk_index]' are used in 'rag/qa.py' without strict isolation. (3) Capability inventory: Ingest, search, and QA tools. (4) Sanitization: Basic HTML tag removal is performed, but there is no filtering for malicious instructions embedded in the data.
Recommendations
- AI detected serious security threats
Audit Metadata