dflow

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 1.00). The skill explicitly configures and instructs AI agents to connect at runtime to a Model Context Protocol server (https://pond.dflow.net/mcp) — e.g., via "claude mcp add" and MCP config entries — which supplies model context that can directly control agent prompts/instructions.

MEDIUM W009: Direct money access capability detected (payment gateways, crypto, banking).

• Direct money access detected (high risk: 1.00). The skill explicitly exposes trading and swap APIs that generate signable transactions (e.g., GET /order returns a base64 transaction, POST /swap returns swapTransaction) and includes concrete code samples that sign with a private key or wallet provider and send transactions to the Solana network. It also documents an "SolanaAgentKit" initialized with a PRIVATE_KEY and states the agent "can execute DFlow swaps." These are specific crypto payment/execution capabilities (wallet signing, transaction submission, platform fee collection, order status polling) — i.e., direct financial execution.