solana-agent-kit

Fail

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: CRITICALEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCREDENTIALS_UNSAFECOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS/REMOTE_CODE_EXECUTION] (CRITICAL): In examples/mcp-server/README.md, the documentation provides a one-liner installation command: curl -fsSL https://raw.githubusercontent.com/sendaifun/solana-mcp/main/scripts/install.sh | bash. Piped remote execution is a critical security risk as the script can be modified at the source to execute arbitrary malicious code on the host system without prior review.
  • [CREDENTIALS_UNSAFE] (HIGH): Multiple files (examples/langchain/README.md, examples/vercel-ai/README.md, examples/mcp-server/README.md) instruct users to store SOLANA_PRIVATE_KEY in environment variables or plaintext JSON configuration files (e.g., claude_desktop_config.json). Private keys in plaintext are susceptible to accidental exposure via logs, process listing, or unauthorized file access.
  • [COMMAND_EXECUTION] (HIGH): The skill provides tools that allow an LLM to perform high-value financial transactions, including transfer, trade, bridge, and most notably signMessage and sendTransaction. While the documentation suggests adding confirmation prompts, the code examples do not implement a programmatic 'Human-in-the-Loop' (HITL) gate at the tool execution level, meaning the agent could be manipulated into signing malicious transactions via prompt injection.
  • [EXTERNAL_DOWNLOADS] (MEDIUM): The skill relies on several external packages from the solana-agent-kit ecosystem and solana-mcp. These sources are not part of the defined 'Trusted Organizations' list and should be audited for supply chain security.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
CRITICAL
Analyzed
Feb 17, 2026, 05:57 PM