Pandadata API

{
  "version": 1,
  "task": {
    "placeholder": "请描述需要查询的数据、市场、标的、日期范围、字段，或贴出待排查的 panda_data 调用与报错",
    "required": true
  },
  "prompt_template": "{{#task}}任务与材料：\n{{task}}\n\n{{/task}}{{#attachments}}用户上传的材料（已放入工作区）：\n{{attachments}}\n\n{{/attachments}}请先读取 references/sdk-0.0.12.md，再从随附 Pandadata 文档中定位准确的 panda_data.get_* 方法及参数/返回字段契约；不得为标记为 not exported 的接口生成 Python SDK 调用，不得臆造方法、字段、标的格式或认证步骤。按文档编写或排查最小可运行调用，附上验证结果，输出中文报告。"
}

Pandadata API

Use this skill to route natural-language data requests to the correct panda_data API method, load the exact parameter/field contract from the bundled Pandadata interface document, and write runnable Python calls.

Workflow

1. Identify the data domain: trading calendar, A-share, futures, options, factors, Hong Kong/US equities, macro, funds, or preferred stocks.
2. Open references/sdk-0.0.12.md, then references/method-index.md, or run ./.venv/bin/python scripts/search_api_docs.py --list-methods when .venv exists.
3. Load the exact method section before coding:

PYTHON_BIN="${PANDADATA_PYTHON:-./.venv/bin/python}"
[ -x "$PYTHON_BIN" ] || PYTHON_BIN=python3

"$PYTHON_BIN" scripts/search_api_docs.py --method get_stock_daily
"$PYTHON_BIN" scripts/search_api_docs.py 股票 分红 --context-lines 4

4. Use the documented method signature and examples from references/api-docs.md. If the index marks a method not exported, do not generate a Python SDK call. Do not invent parameters, field names, symbols, or authentication steps.
5. For real API calls, prefer scripts/call_api.py; it loads saved credentials, runs setup when credentials are missing, initializes panda_data in the same process, then calls the API.

Calling Pattern

Use import panda_data; this skill requires panda_data==0.0.12 and Python 3.10 or newer. Pandadata calls return DataFrame-like tabular results in the examples.

Run SDK commands with an interpreter that has 0.0.12 installed. Set PANDADATA_PYTHON when needed; common virtual-environment paths are .venv/bin/python on POSIX and .venv\Scripts\python.exe on Windows. The runtime scripts reject any other installed SDK version.

Install the runtime SDK when real API calls are required:

PYTHON_BIN="${PANDADATA_PYTHON:-./.venv/bin/python}"
[ -x "$PYTHON_BIN" ] || PYTHON_BIN=python3
"$PYTHON_BIN" -m pip install -r requirements.txt

For first-time setup, prefer the interactive setup script. It installs the pinned SDK, verifies version 0.0.12, prompts for credentials without echoing the password, validates login, and can optionally save a shell env file at ~/.pandadata/pandadata.env.

PYTHON_BIN="${PANDADATA_PYTHON:-./.venv/bin/python}"
[ -x "$PYTHON_BIN" ] || PYTHON_BIN=python3
"$PYTHON_BIN" scripts/setup_runtime.py

Initialize the SDK before any get_* call. The installed SDK raises ClientNotInitializedError until init_token() succeeds.

import panda_data
import os

panda_data.init_token(
    username="YOUR_USERNAME",
    password="YOUR_PASSWORD",
    base_url=os.environ["JAVA_SERVICE_BASE_URL"],
)

The SDK can also read DEFAULT_USERNAME, DEFAULT_PASSWORD, and JAVA_SERVICE_BASE_URL from environment variables. Pass the plain password; the SDK hashes it internally. Do not assume a fixed service URL; use the configured environment or the saved env file. SDK 0.0.12 keeps the token in memory and persists encrypted credentials in user.json when init_token() succeeds; --no-save-env only disables the separate plaintext shell env file.

If credentials were saved by scripts/setup_runtime.py, load them before running data scripts:

source ~/.pandadata/pandadata.env

Preferred CLI runner for agents:

PYTHON_BIN="${PANDADATA_PYTHON:-./.venv/bin/python}"
[ -x "$PYTHON_BIN" ] || PYTHON_BIN=python3
"$PYTHON_BIN" scripts/call_api.py \
  --method get_stock_daily \
  --params '{"symbol":["000001.SZ"],"start_date":"20250101","end_date":"20250131","fields":[],"indicator":"000300","st":true}'

call_api.py behavior:

• Load credentials from current env or ~/.pandadata/pandadata.env.
• If credentials are missing, run scripts/setup_runtime.py interactively.
• Call panda_data.init_token() in the same Python process.
• Invoke the requested panda_data.get_* method and emit JSON by default.

For custom Python scripts, import the runtime helper so the login check happens in the same process as the API call:

from pathlib import Path
import sys

sys.path.append(str(Path("scripts").resolve()))
from pandadata_runtime import init_pandadata

panda_data = init_pandadata()
result = panda_data.get_stock_daily(
    symbol=["000001.SZ"],
    start_date="20250101",
    end_date="20250131",
    fields=[],
)
print(result)

import panda_data

result = panda_data.get_stock_daily(
    symbol=["000001.SZ"],
    start_date="20250101",
    end_date="20250131",
    fields=[],
    indicator="000300",
    st=True,
)
print(result)

Before running real calls in a new environment:

python - <<'PY'
from importlib.metadata import version
import panda_data
print("panda_data import ok", version("panda-data"), panda_data.__file__)
PY

If panda_data is unavailable or credentials are missing, report that the SDK/runtime is not configured. The supplied interface document does not define installation, token, or login setup.

PyPI runtime dependencies from panda_data==0.0.12: pandas>=2.0.0, numpy>=1.22,<2.0, python-snappy>=0.7.3, python-dotenv>=1.0.0, PyYAML>=6.0, zstandard>=0.22.0, duckdb, pyarrow, websockets>=13.0, and requests.

Core Conventions

• Dates in the document use YYYYMMDD strings, for example 20250131.
• A-share symbols use suffix format such as 000001.SZ or 600000.SH.
• Exchange codes documented for calendars include SH, HK, and US.
• Many APIs accept fields=[] for all fields, but some tables describe fields as string; follow the exact method example when present.
• Some APIs accept list inputs such as symbol=["000001.SZ"]; others show scalar strings such as symbol="000001.SZ". Match the target method's own example.
• For broad/unfiltered calls, warn the user when the API may return a large table.

Limitations

• Live availability depends on the configured Pandadata service, account permissions, and upstream data coverage.
• Treat bundled interface documentation as a contract reference, but apply the SDK 0.0.12 availability exceptions in references/sdk-0.0.12.md.
• This skill is data-access and research tooling. Its examples and outputs do not constitute investment advice.

Reference Files

• references/method-index.md: compact method map grouped by domain, with line numbers into api-docs.md.
• references/sdk-0.0.12.md: supported SDK contract, authentication changes, method reconciliation, and known upstream limitations.
• references/api_catalog.json: method-to-gateway endpoint mapping for API routing checks, preserving the MCP gateway's /pandaData/... dispatch paths.
• references/api-docs.md: full Chinese Pandadata interface reference reconciled with SDK 0.0.12.
• references/agent-integration.md: installation, loading, and smoke-test patterns for Claude Code, Codex, Hermes, OpenClaw, Cursor, and WorkBuddy.

Agent Usage Rules

• Prefer the bundled reference over memory. Search before answering API-specific questions.
• Quote method names and parameters exactly as documented.
• Run scripts/call_api.py --method <method> --params '{}' --dry-run when validating that SDK 0.0.12 exports a method.
• Keep examples minimal and executable; use head(), shape, or explicit row counts when validating.
• Separate data retrieval from analysis: first obtain/verify the DataFrame, then transform or analyze it.
• If a method returns empty data, check date range, symbol format, and required filters before assuming service failure.