---

name: didi-ride description: 通过滴滴出行 MCP 打车、查价、查询订单、取消订单、规划路线（驾车/公交/步行/骑行）和搜索地点。Use when the user mentions 滴滴, 打车, 叫车, 回家/上班要打车, 查一下从 A 到 B 多少钱/怎么走, 查询订单, 司机在哪/多久到, 预约叫车, 路线规划, DiDi, ride-hailing, or booking a taxi. when_to_use: | Trigger for anything involving the user's DiDi (滴滴出行) account: book a ride ("打车去…", "回家", "上班"), get a fare/route estimate, query an existing order (driver location, ETA, trip progress), cancel an order, plan a driving/transit/walking/cycling route, or search places / reverse-geocode. Creating and cancelling orders act on real money and a real driver, so those writes are gated behind explicit confirmation. connections: [didi] allowed_tools: [Bash] license: Apache-2.0 metadata: author: acedatacloud version: "1.0"

DiDi Ride (滴滴出行)

Drive the user's DiDi (滴滴出行) account through the DiDi MCP server: book rides, estimate fares, track orders, cancel, and plan routes.

The didi BYOC connector injects one env var into the sandbox:

• DIDI_MCP_KEY — the user's DiDi MCP key. Secret — never echo, print, or log it.

If DIDI_MCP_KEY is missing, tell the user to connect the DiDi connector at studio.acedata.cloud/console/connectors (they get the key by scanning the QR in the 滴滴出行 App or via https://mcp.didichuxing.com/claw).

CLI

The skill ships a stdlib-only helper that speaks the MCP Streamable-HTTP protocol to DiDi. Run this resolver at the top of every Bash block below — each Bash call is a fresh shell, and $SKILL_DIR points at the LAST skill loaded this turn, so anchor on our own script:

DIDI="$SKILL_DIR/scripts/didi.py"; [ -f "$DIDI" ] || DIDI=$(find /tmp -maxdepth 8 -path '*/skills/*/scripts/didi.py' 2>/dev/null | head -1)
[ -f "$DIDI" ] || { echo "didi-ride script not found (SKILL_DIR=$SKILL_DIR)" >&2; exit 1; }

Two commands:

python3 $DIDI list                        # list tools + their JSON input schemas
python3 $DIDI call <tool> '<json-args>'   # call any tool

Always check references/api_references.md for the exact tool + parameter names before calling. When unsure, run python3 $DIDI list — it returns the authoritative input schema for every tool straight from DiDi. Do not guess parameter names (common mistakes: keyword → keywords, region → city, four coord fields → six from_name/from_lat/from_lng/to_name/to_lat/to_lng).

All argument values must be strings (including coordinates and product_category), e.g. {"product_category":"1"} not {"product_category":1}.

Write gating (real money / real driver)

taxi_create_order and taxi_cancel_order are gated: without a trailing --confirm the helper only DRY-RUNS and changes nothing. Run once without --confirm to preview, then re-run with --confirm as the last argument after the user approves:

python3 $DIDI call taxi_create_order '{"estimate_trace_id":"...","product_category":"1"}' --confirm

• Even when the user says "打车" / "取消订单", confirm the concrete details (起终点、车型 for booking; the order for cancelling) before adding --confirm.
• Cancel intent ≠ cancel confirmation — always ask "确认取消吗？" first.

Booking flow (最小可执行)

1. Resolve addresses — maps_textsearch (never invent coordinates; don't reuse coordinates from earlier turns — the user may have moved).
   • If the user references an address alias (家 / 公司 / etc.) that you don't have, ask them; this skill has no stored preferences.
2. Confirm start/end — if maps_textsearch returns ≥2 candidates, list at least the top 3 and let the user pick; a single exact match needs no confirmation.
3. Estimate — taxi_estimate; record the returned traceId / estimate_trace_id. It expires (-32021) — re-estimate if stale.
4. Pick car type — user's current message wins ("叫快车"→product_category 1, "专车"→8); otherwise ask. Only use categories present in the taxi_estimate response; never silently substitute a different service level (快车 1 ≠ 特惠快车 201).
5. Create order — taxi_create_order with the latest estimate_trace_id (dry-run → confirm → --confirm).
6. Report — order id, start/end, car type, estimated fare. Tell the user they can send 「查询订单」 to check status.

Query an order

Order id comes from (in priority): the user's message → the most recent order created this conversation → else ask.

python3 $DIDI call taxi_query_order '{"order_id":"ORDER_ID"}'

Status codes (code):

code	含义	输出
0	匹配中	⏳ 正在为您匹配司机
1	司机已接单	展示司机姓名、车型、车牌、电话、距离与预计到达时间
2	司机已到达	🔔 司机已到达上车点
4	行程进行中	🚗 行程已开始
5	订单完成	✅ 行程结束（展示费用，如有）
6	系统取消	❌ 订单已被系统取消
7	已取消	❌ 订单已取消
3 / 8-12	其他终态	显示对应状态描述

Routes & places (no order needed)

python3 $DIDI call maps_direction_driving  '{"from_name":"...","from_lat":"...","from_lng":"...","to_name":"...","to_lat":"...","to_lng":"..."}'
python3 $DIDI call maps_direction_transit  '{...}'   # 公交
python3 $DIDI call maps_direction_walking  '{...}'   # 步行
python3 $DIDI call maps_direction_bicycling '{...}'  # 骑行
python3 $DIDI call maps_place_around '{"keywords":"咖啡","lat":"...","lng":"..."}'
python3 $DIDI call maps_regeocode '{"lat":"...","lng":"..."}'

Gotchas

• Never print the key or the raw endpoint URL — the helper handles the transport and keeps the key internal.
• taxi_create_order takes only estimate_trace_id, product_category, and optional caller_car_phone. Don't pass the estimate's coordinate/name fields.
• No stored preferences: this skill doesn't remember home/work/car type — ask the user, or wire richer memory in a higher layer.
• Auth failure surfaces as -32002 (or HTTP 401/403): the key is missing/expired → have the user reconnect the connector.

Setup: See authentication. This skill uses the didi connector's injected DIDI_MCP_KEY, not ACEDATACLOUD_API_TOKEN.