didi-ride

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The skill ingests outsider-authored free text at runtime when the user provides search/place/route text (e.g., mapped into maps_textsearch.keywords / maps_place_around.keywords) which the agent then sends to the DiDi MCP; this is direct user-controlled text used without pre-selecting a specific trusted item.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill's runtime helper (scripts/didi.py) sends requests to the DiDi MCP server at https://mcp.didichuxing.com/mcp-servers to fetch tool schemas and perform calls (including state-changing rides), which directly controls the agent's tool parameters and behavior and is required for operation.

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

• Direct money access detected (high risk: 1.00). The skill provides definitive, specific APIs to create and cancel real DiDi ride orders (taxi_create_order / taxi_cancel_order) that "act on real money and a real driver" and exposes a connector key (DIDI_MCP_KEY) allowing the agent to act on the user's DiDi account. Although writes are gated behind an explicit --confirm flow, the skill explicitly enables creating/cancelling paid transactions via the DiDi MCP tools and example CLI calls, which constitutes direct financial execution authority for ride purchases.