功夫财经 | KungfuFinance#

Publisher#

• Organization: kungfu-trader
• Source repository: github.com/kungfu-trader/kungfu-skills
• Tianshan API operator: kungfu-trader (same organization)

Runtime Model#

• Local runtime: Node.js .mjs scripts (bundled executable scripts with zero npm dependencies — no npm install needed)

• Outbound APIs:

  • https://tianshan-api.kungfu-trader.com — deterministic products (snapshot, finance, bucket, strategy, researcher, bayesian monitor)
  • https://push2.eastmoney.com / https://push2his.eastmoney.com — stock/sector deep research market data (free public API, no key required)
  • https://ifzq.gtimg.cn / https://qt.gtimg.cn — fallback market data via Tencent Finance (free public API, no key required)
  • https://wry-manatee-359.convex.site / https://clawhub.ai — check-update version query (ClawHub registry, public, no key required)

• Authentication: KUNGFU_OPENKEY environment variable, sent as Authorization: Bearer <token> (Tianshan API only; EastMoney/Tencent/ClawHub APIs require no authentication)

• Platform header: KUNGFU_PLATFORM (optional, defaults to openclaw)

• File I/O: reads bundled assets; indicator-chart flow writes SVG/PNG to ~/.openclaw/workspace/finance-master/charts/

• Subprocesses: indicator-chart flow invokes inkscape for SVG→PNG conversion; no other subprocesses

All environment variables are read from the host process; none are accepted from user prompts.

Optional Research Search Surface#

When separately configured, stock/sector deep research may call an additional search endpoint. This surface is independent from the Tianshan API and uses its own credential boundary:

Env VarPurposeKUNGFU_ENABLE_RESEARCH_SEARCHSet to 1 to enableKUNGFU_RESEARCH_SEARCH_PROVIDERProvider name (currently web_search)KUNGFU_RESEARCH_SEARCH_ENDPOINTSearch service URLKUNGFU_RESEARCH_SEARCH_API_KEYSearch service credentialKUNGFU_RESEARCH_SEARCH_TIMEOUT_MSSearch request timeout (default: 15000)

When search is enabled, requests are POSTed as JSON to KUNGFU_RESEARCH_SEARCH_ENDPOINT with KUNGFU_RESEARCH_SEARCH_API_KEY as Bearer token. KUNGFU_OPENKEY is never reused for search requests. If search is enabled but misconfigured, the result returns misconfigured status instead of silently failing.

Non-Secret Tuning Variables#

Env VarPurposeKUNGFU_ENABLE_EXPERIMENTAL_PRODUCTSSet to 1 to enable unreleased productsKUNGFU_RESEARCH_DEFAULT_TARGET_DATEOverride default research date (YYYYMMDD format, for testing)

Rollout Status#

• Deterministic products (snapshot, finance, bucket, strategy, researcher, bayesian monitor) use revalidated Tianshan backend routes
• Preview stock-research / sector-research use public EastMoney/Tencent APIs for market data — no backend dependency for these flows
• Experimental products require KUNGFU_ENABLE_EXPERIMENTAL_PRODUCTS=1
• Preview stock-research / sector-research produce markdown_svg_preview with explicit degradation reporting

Data Handling & Network Surface#

• KUNGFU_OPENKEY is sent only to tianshan-api.kungfu-trader.com — never to EastMoney, Tencent, or any other host
• EastMoney and Tencent Finance APIs are public and require no authentication
• The code contacts no other hosts beyond those listed in the Runtime Model table above
• Do not use this skill with secrets, personal data, or other sensitive user content

Scope#

• Mainland China A-shares only
• Direct market-data lookup stays within one stock or one sector
• No arbitrary whole-market screening
• Not for US stocks
• Not for Hong Kong stocks
• Not for cryptocurrencies
• Not for futures or forex
• No free-form SQL
• No user-facing base URL
• No user-facing runtime credential parameters

Public Intents#

Route the user into the currently enabled intent families below. Do not expose internal product names unless needed for implementation.

1. Stock Snapshot#

Use when the user wants one A-share stock's current state, recent走势, K-line window, holders,筹码分布,支撑压力位, or other narrow deterministic data.

Default behavior:

• Prefer the narrowest data product that answers the question
• If the stock code is unknown, prefer instrument_name
• If code mode is used, exchange_id must be SSE or SZE

2. Finance Analysis#

Use when the user wants one A-share stock's基础财务指标或基础财务上下文。

Default behavior:

• Prefer finance_basic_indicators for narrow factual questions
• Prefer finance_context for lightweight finance context

3. Sector Analysis#

Use when the user asks about one A-share industry or sector.

Default behavior:

• deterministic sector detail products may still use sector_name
• fuzzy sector resolution is not exposed as a standalone public product, but sector-research resolves theme words by matching against the full EastMoney concept sector list
• preview sector-research accepts sector_name or sector_id directly; when the local resolver cannot disambiguate, it returns needs_input for the user to provide a BK code

3C. Sector Deep Research#

Use when the user wants one A-share sector's deep research, hype-cycle view, thematic thesis, dragon-head observation, committee-style bull/bear review, or a longer structured report.

Default behavior:

• current runtime only supports one mainland China A-share sector per invocation
• the preview path accepts sector_name or sector_id; target_date is optional and defaults to the current market date in Asia/Shanghai
• the current version is markdown_svg_preview and returns explicit degradations for remaining search / adapter gaps
• the result surface now also exposes the migrated orchestration contract from the original sector-analysis skill
• the result surface now includes report_svg and quality_gate
• structured inputs now use public EastMoney APIs: concept sector list resolution, sector constituents, sector index K-line, and per-leader stock quotes and money flow
• the sector resolver matches user input against the full EastMoney concept sector list; if ambiguous, returns needs_input for the user to provide a BK code
• when the separate web_search runtime is enabled and configured, macro / policy / catalyst / industry-trend / competition / money-flow evidence may be attached as external search evidence (7 search buckets)
• otherwise the report must surface missing search evidence as degraded coverage instead of silently inventing facts

3B. Stock Deep Research#

Use when the user wants one A-share stock's deep research, investment thesis, catalyst map, committee-style bull/bear review, or a longer structured report.

Default behavior:

• current runtime only supports one mainland China A-share stock per invocation
• the preview path accepts optional target_date and defaults it to the current market date in Asia/Shanghai
• the current version is markdown_svg_preview and returns explicit degradations for remaining search / adapter gaps
• the result surface now also exposes the migrated orchestration contract from the original stock-analysis-v2 skill
• the result surface now includes report_svg and quality_gate
• structured market data now comes from public EastMoney APIs: real-time quote, daily K-line, real weekly K-line (klt=102, not aggregated from daily), and 20-day money flow; Tencent Finance APIs serve as fallback
• financial statements, company profile, and other non-market data are sourced from web search (8 search buckets: macro, catalyst, risk, competition, financials, company profile, governance, sector trend)
• when the separate web_search runtime is enabled and configured, these evidence buckets are populated; otherwise the report surfaces explicit degradations

4. Bucket Flow#

Use when the user wants to:

• view current-user buckets
• choose one existing bucket
• add one or more instruments into that bucket

Default behavior:

• only operate on the current authenticated user's own buckets
• do not create or delete buckets in the current release build
• when adding a batch, resolve instruments first and filter invalid ones before writing
• when required information is missing or ambiguous, return a structured needs_input response so the host can continue the dialogue
• if all inputs are invalid, return needs_input for corrected instruments instead of calling the write API

5. Strategy Flow#

Use when the user wants to:

• view public strategies
• view public strategies by market mode
• view current-user private strategies
• query one public or private strategy on one stock
• query one public or private select strategy's whole-market picks for one day or one date range
• count buy signals for one stock across eligible public paid strategies
• batch scan one public paid strategy across multiple stocks

Default behavior:

• strategy module requires runtime auth even when listing public strategies
• public and private strategies both support single-strategy query
• public and private select strategies both support controlled whole-market result query
• private strategies do not support first-stage batch scan
• batch scan only supports public strategies with non-empty lago_plan
• market-select supports InstrumentSelect, TemplateInstrumentSelect, BuySell, and TemplateBuySell
• when market-select is used on BuySell or TemplateBuySell, whole-market results only return buy points
• market-select accepts either target_date, or strategy_start_date + strategy_end_date
• if market-select mixes the two date modes, or only provides half of a range, return a structured needs_input response
• when strategy selector or stock input is missing or ambiguous, return a structured needs_input response so the host can continue the dialogue
• when the selected public paid strategy lacks permission for batch scan, return a structured blocked response instead of pretending the scan succeeded

6. Researcher Flow#

Use when the user wants to:

• view top researchers by score
• view one stock's covered researchers together with report titles
• view one researcher's reports

Default behavior:

• researcher module requires runtime auth
• researcher backend also enforces Journeyman membership or above
• stock-reports is a composed flow built from researcher score plus researcher report list
• author-reports prefers researcher-author-id
• if the user only gives researcher-name, resolve it conservatively; if not unique, return needs_input
• do not pretend the backend supports direct global researcher-name search if it does not

7. Bayesian Monitor Flow#

Use when the user wants to:

• view current-user bayesian monitor tasks
• choose one existing bayesian monitor task
• inspect that task's initial report and recent monitor records

Default behavior:

• bayesian monitor module requires runtime auth
• only operate on the current authenticated user's own tasks
• current release build is read-only: do not create, delete, or run bayesian monitor tasks
• list returns task summaries only and must not expose full original_report
• reports supports bayesian-task-id and conservative bayesian-topic resolution
• when task selector is missing or ambiguous, return a structured needs_input response so the host can continue the dialogue
• when the current user has no tasks, return a structured blocked response instead of pretending the task exists

8. Indicator Chart#

Use when the user wants to see one A-share stock's technical indicator chart. The script automatically renders SVG, converts to PNG via inkscape, and returns file paths (no SVG text in stdout). Supported chart types:

• kline — K线图 (candlestick chart with optional chip distribution overlay, price level lines, and custom annotations)
• lushan_shadow — 庐山照影图 (CM mountain + FC shadow dual-bar chart)
• lushan_4season — 庐山四季图 (GG/HY/DP lines + RS bars + bottom resonance/strong markers)
• shuanglun — 双轮驱动图 (K-wheel + R-wheel stacked bars with trend coloring)
• liumai — 六脉神剑图 (6-signal heatmap matrix with buy/sell signal row)
• smart_money — 聪明钱图 (smart money vs follow money dual-line chart)
• finance_score — 个股综合得分 (radar chart with 行业统治力/财务健康度/发展前景 three-dimensional scoring)

K-line chart additional options:

• --with-chip-peak true — overlay chip distribution (筹码峰) on the right side
• --with-price-levels true — draw support/resistance/concentration price level lines (阻力位/支撑位)
• --annotations '[{"date":"20260328","price":15.5,"type":"buy","label":"买入点","color":"#ef4444"}]' — mark custom points on the chart

9. Finance Score#

Use when the user wants one A-share stock's comprehensive scoring across three dimensions:

• 行业统治力 (Industry Dominance) — scored 0-5 with S/A/B/C/D/E grade
• 财务健康度 (Financial Health) — scored 0-5 with S/A/B/C/D/E grade
• 发展前景 (Growth Potential) — scored 0-5 with S/A/B/C/D/E grade
• 综合得分 (Total Score) — 0-15

Available as both a data product ( finance_score) and a rendered SVG chart ( indicator-chart --chart-type finance_score).

10. Subscription Flow (自选)#

Use when the user wants to:

• view current-user subscribed instruments (自选列表)
• add instruments to subscription (加自选)
• remove instruments from subscription (删自选)
• view anomaly report for one subscribed instrument (异动报告)
• view aggregate anomaly summary (异动汇总)

Default behavior:

• only operate on the current authenticated user's own subscriptions
• when adding, resolve instrument names first and skip invalid or duplicate entries
• list groups subscriptions by instrument and shows instrument name, code, auto-update status
• anomaly returns unusual movements sorted by date descending, plus signals count and analysis
• summary returns aggregate UM_ct (异动数量) and UM_SG_ct (异动信号) counts
• when required information is missing, return a structured needs_input response

Not Yet Enabled In Experience Build#

The following capabilities are still under route-contract cleanup and should not be used in the current public rollout:

• full finance-analysis bundles
• money flow
• fuzzy sector resolution / similar sectors
• market news lookup
• standalone fuzzy sector resolution as a separate public deterministic intent outside sector-research

Intent Routing#

Pick one default route from the list below.

• User asks "这只股票现在怎么样 / 最近走势 / 看看价格": Route to Stock Snapshot
• User asks "分析财报 / 财务怎么样 / 估值怎么看": Route to Finance Analysis
• User asks "分析白酒 / 算力 / 某个板块 / 某个概念": Route to Sector Analysis
• User asks "给我做贵州茅台的深度研究 / 做一个投资论题 / 看催化剂和风险 / 写一份个股深度报告": Route to Stock Deep Research
• User asks "看看我的票池 / 把这些股票加到票池里": Route to Bucket Flow
• User asks "有哪些策略 / 看看我的私人策略 / 某个策略今天有没有买卖点 / 庐山升龙今天选了哪些股票 / 某个策略最近一周选股结果 / 用某个策略扫一组股票": Route to Strategy Flow
• User asks "画一下贵州茅台的K线图 / 庐山照影 / 庐山四季 / 双轮驱动 / 六脉神剑 / 聪明钱 / 带筹码峰的K线 / 标注买卖点": Route to Indicator Chart
• User asks "贵州茅台的综合得分 / 行业统治力 / 财务健康度 / 发展前景": Route to Finance Score
• User asks "评分最高的研究员有哪些 / 某个股票有哪些研究员和研报 / 某个研究员写过哪些研报": Route to Researcher Flow
• User asks "看看我的贝叶斯监控 / 某个贝叶斯监控任务的报告 / 某个主题最近的贝叶斯监控记录": Route to Bayesian Monitor Flow
• User asks "功夫财经连接正常吗 / OpenKey 配置对吗 / health check / 检查配置": Route to Health Check
• User asks "配置 OpenKey / 设置 OpenKey / 我的 key 是 kf_xxx": Route to Config OpenKey
• User asks "看看我的自选 / 自选列表 / 我订阅了哪些股票": Route to Subscription Flow (action: list)
• User asks "把贵州茅台加入自选 / 加自选 / 订阅这些标的": Route to Subscription Flow (action: add)
• User asks "从自选中移除平安银行 / 取消订阅 / 删除自选": Route to Subscription Flow (action: delete)
• User asks "贵州茅台有什么异动 / 查看异动 / 最近的异动报告": Route to Subscription Flow (action: anomaly)
• User asks "自选整体异动情况 / 今天有多少异动 / 异动汇总": Route to Subscription Flow (action: summary)
• User asks "功夫财经有更新吗 / 检查更新 / check update / 升级 skill": Route to Check Update

If the user asks for arbitrary whole-market screening or market-news bundles, do not use this skill in the current release build.

Input Rules#

Stock Identification#

For single-stock requests, use exactly one:

• --instrument-name
• --instrument-id and --exchange-id

Never mix both modes. If the stock code is unknown, prefer --instrument-name. When using code mode, --exchange-id must be SSE or SZE.

Stock Date Mode#

For stock products and flows, use:

• --target-date
• optional --visual-days-len

Strategy Date Mode#

For strategy actions, use:

• signal and count: --target-date with optional --visual-days-len
• batch-scan: --target-date
• market-select: either --target-date, or --strategy-start-date plus --strategy-end-date

Never mix single-day mode and range mode in the same market-select request.

Sector Mode#

• For sector detail products, use exactly one:
  • --sector-name
  • --sector-id

Prefer --sector-name. Only reuse --sector-id after the backend has already returned it. The current release build does not expose fuzzy sector resolution as a standalone public data product, but sector-research can resolve theme words via the EastMoney concept sector list.

Hidden Parameters#

Do not generate or expose these from the model side:

• start_date
• end_date
• is_realtime
• limit
• days
• periods

Internal Building Blocks#

The internal deterministic product contract lives here:

• data_products.md

Use that file only when you need exact internal product names, allowed parameter shapes, or output keys. Do not dump that whole catalog into the user-facing reasoning path by default.

Bucket flow contract lives here:

• bucket_flow.md

Strategy flow contract lives here:

• strategy_flow.md

Researcher flow contract lives here:

• researcher_flow.md

Bayesian monitor flow contract lives here:

• bayesian_monitor_flow.md

Research Methodologies#

Documented research methods live here:

• README.md
• runtime_parity.md
• stock-analysis
• sector-analysis

Use them as methodology references and migration inputs for deep research. The repo build currently has preview CLI/runtime entrypoints via stock-research and sector-research. The sector preview accepts sector_name first and uses best-effort EastMoney concept list resolution before falling back to needs_input. Read runtime_parity.md before treating any research wrapper or asset as implementation-ready truth.

Prompt assets are flow-specific. Do not mix finance-analysis formatting requirements into movement-analysis outputs, or vice versa.

Standard Commands#

Stock Snapshot#

bash

node scripts/flows/run_data_request.mjs --product price_snapshot --instrument-name 贵州茅台 --target-date 20260301

Sector Detail#

bash

node scripts/flows/run_data_request.mjs --product sector_performance --sector-name 白酒 --target-date 20260301

Bucket List#

bash

node scripts/router/run_router.mjs bucket --bucket-action list

Bucket Add#

bash

node scripts/router/run_router.mjs bucket --bucket-action add --bucket-name 观察池 --bucket-instrument 贵州茅台 --bucket-instrument 平安银行

Strategy List#

bash

node scripts/router/run_router.mjs strategy --strategy-action list

Strategy Signal#

bash

node scripts/router/run_router.mjs strategy --strategy-action signal --strategy-name 三分归元 --instrument-name 贵州茅台 --target-date 20260331 --visual-days-len 22

Strategy Batch Scan#

bash

node scripts/router/run_router.mjs strategy --strategy-action batch-scan --strategy-id 900 --strategy-instrument 600519.SSE --strategy-instrument 000001.SZE --target-date 20260331

Strategy Market Select#

bash

node scripts/router/run_router.mjs strategy --strategy-action market-select --strategy-name 庐山升龙 --target-date 20260331

Stock Deep Research#

bash

node scripts/router/run_router.mjs stock-research --instrument-name 贵州茅台 --target-date 20260331 --visual-days-len 120

Sector Deep Research#

bash

node scripts/router/run_router.mjs sector-research --sector-name 机器人 --target-date 20260331 --visual-days-len 780

Indicator Chart — K-line#

bash

node scripts/router/run_router.mjs indicator-chart --chart-type kline --instrument-name 贵州茅台 --target-date 20260331 --visual-days-len 120

Indicator Chart — K-line with Chip Peak and Price Levels#

bash

node scripts/router/run_router.mjs indicator-chart --chart-type kline --instrument-name 贵州茅台 --target-date 20260331 --visual-days-len 120 --with-chip-peak true --with-price-levels true

Indicator Chart — K-line with Annotations#

bash

node scripts/router/run_router.mjs indicator-chart --chart-type kline --instrument-name 贵州茅台 --target-date 20260331 --visual-days-len 60 --annotations '[{"date":"20260310","type":"buy","label":"买入"},{"date":"20260325","type":"sell","label":"卖出"}]'

Indicator Chart — 庐山照影图#

bash

node scripts/router/run_router.mjs indicator-chart --chart-type lushan_shadow --instrument-name 贵州茅台 --target-date 20260331 --visual-days-len 120

Indicator Chart — 庐山四季图#

bash

node scripts/router/run_router.mjs indicator-chart --chart-type lushan_4season --instrument-name 贵州茅台 --target-date 20260331 --visual-days-len 120

Indicator Chart — 双轮驱动图#

bash

node scripts/router/run_router.mjs indicator-chart --chart-type shuanglun --instrument-name 贵州茅台 --target-date 20260331 --visual-days-len 120

Indicator Chart — 六脉神剑图#

bash

node scripts/router/run_router.mjs indicator-chart --chart-type liumai --instrument-name 贵州茅台 --target-date 20260331 --visual-days-len 120

Indicator Chart — 聪明钱图#

bash

node scripts/router/run_router.mjs indicator-chart --chart-type smart_money --instrument-name 贵州茅台 --target-date 20260331 --visual-days-len 120

Finance Score (data only)#

bash

node scripts/flows/run_data_request.mjs --product finance_score --instrument-name 贵州茅台 --target-date 20260331

Finance Score (SVG radar chart)#

bash

node scripts/router/run_router.mjs indicator-chart --chart-type finance_score --instrument-name 贵州茅台 --target-date 20260331

Researcher Rank#

bash

node scripts/router/run_router.mjs researcher --researcher-action rank --researcher-rank-by 1m --researcher-limit 10

Researcher Stock Reports#

bash

node scripts/router/run_router.mjs researcher --researcher-action stock-reports --instrument-name 贵州茅台

Researcher Author Reports#

bash

node scripts/router/run_router.mjs researcher --researcher-action author-reports --researcher-author-id author-1

Health Check#

bash

node scripts/router/run_router.mjs health

Returns { healthy: true/false, checks: { openkey, api_test } }. Verifies KUNGFU_OPENKEY is set and authenticates against the backend.

Config OpenKey#

bash

node scripts/router/run_router.mjs config-openkey --openkey kf_ok_live_xxxxx

Validates the key against the backend, then writes it to ~/.openclaw/.env. Returns user_id and plan_code on success.

Subscription List (自选列表)#

bash

node scripts/router/run_router.mjs subscription --subscription-action list

Subscription Add (加自选)#

bash

node scripts/router/run_router.mjs subscription --subscription-action add --subscription-instrument 贵州茅台 --subscription-instrument 平安银行

Subscription Delete (删自选)#

bash

node scripts/router/run_router.mjs subscription --subscription-action delete --subscription-instrument 贵州茅台

Subscription Anomaly Report (异动报告)#

bash

node scripts/router/run_router.mjs subscription --subscription-action anomaly --instrument-name 贵州茅台 --target-date 20260401

Subscription Summary (异动汇总)#

bash

node scripts/router/run_router.mjs subscription --subscription-action summary

Check Update#

bash

node scripts/router/run_router.mjs check-update

Checks ClawHub registry for a newer version. Returns { update_available, local_version, remote_version, update_command }. When an update is available, the agent should run the returned update_command ( clawhub update kungfu-finance) in a shell to apply it.