by simonlin1212
全球宏观预期概率面板 (AI Skill) — Polymarket+Kalshi 双源 · 5端点 · 10模块 · 零鉴权 | Global-macro-probability panel skill for AI coding assistants (prediction-market data pipeline)
Unlocks once the catalog security scan passes (runs nightly).
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
# Add to your Claude Code skills
git clone https://github.com/simonlin1212/globalpercentGuides for using ai agents skills like globalpercent.
📦 项目主页：https://github.com/simonlin1212/globalpercent — 更新、反馈、支持作者
作者：Simon 林 · 抖音「Simon林」· 公众号「硅基世纪」
This skill packages everything needed to build a panel that reads prediction-market probabilities from Polymarket + Kalshi (both public, read-only, no account) and presents them as a macro sentiment thermometer grouped by module.
It is NOT a single API call. The value is in the system: classification, dedup, multi-source merge, snapshot caching, async refresh, translation, and a dozen non-obvious gotchas that each cost real debugging time. All of that is captured here so you don't rediscover it.
A prediction market's price IS a probability. A contract trading at $0.62 means the market collectively bets 62% the event happens. Reading those numbers is free, needs no wallet or account — only trading does. So you can pull a global, money-backed sentiment read on Fed decisions, geopolitics, AI milestones, etc. for zero cost.
Use it as a thermometer, not a trading tool. 70–84% of prediction-market traders lose money; the edge belongs to HFT market-makers, not "AI that predicts well." Frame the panel as "what is the market's mood?", never as a buy/sell signal.
Polymarket Gamma API ─┐
├─► per-source fetchers shape every market into ONE common schema
Kalshi events API ────┘ (question, prob_yes, change_24h, volume_24h, source, …)
│
shared taxonomy classifies each into a module
(货币政策/宏观/地缘/政治/股指大宗/AI + reference group)
│
aggregator: merge → group by module → cap floods →
translate titles → pin a disk SNAPSHOT
│
/pulse/overview (instant from snapshot; refresh = async rebuild)
│
React panel: module sections + source badges + trend chart
Follow in order. Read the two reference files first — they hold the details this overview compresses.
reference/apis.md — exact endpoints, field names, and every API gotcha
(Kalshi's *_dollars field rename, no volume-sort, broken category filter, etc.).
Verify the endpoints live before coding — these APIs change.reference/architecture.md — the design decisions and the hard-won gotchas
(async refresh, translation rate-limiting, empty-pull-never-overwrite, multi-leg events).code/backend/ into your stack:
market_taxonomy.py — module list + keyword classifier (source-agnostic). Tune keywords.polymarket_signals.py — Polymarket Gamma/CLOB fetch + shape + snapshot.kalshi_signals.py — Kalshi events fetch + shape + snapshot (note the retry/empty-guard).polymarket_translate.py — optional title translation (swap in your LLM client).market_pulse.py — the aggregator + async-refresh snapshot model. This is the core.api_routes.py — FastAPI routes (/pulse/overview, /polymarket/history).code/frontend/:
EventProbabilityPanel.tsx — module sections, source badges, EN-primary/CN-secondary
titles, multi-leg "档位 (market line)" chips, collapsed reference group, async-refresh poll.ProbabilityTrend.tsx — echarts line chart (swap for your chart lib if needed)./pulse/overview?refresh=true once to build the first snapshot
(can take minutes — Kalshi is slow), then confirm normal loads are instant and modules
are populated and capped.The reference code came from a working FastAPI + React dashboard. Three things are environment-specific — find and replace them:
| In the code | What it is | Swap for |
|---|---|---|
from src.config.paths import get_data_dir |
where snapshots/cache live (has a ~/.vibe-trading fallback) |
your app's data dir |
from src.providers.llm import build_llm (in polymarket_translate.py) |
the LLM used to translate titles to Chinese | your LLM client, or drop translation entirely (English-only) |
ProbabilityTrend, Tailwind classes, lucide icons, @/ alias |
UI stack | your component/design system |
Everything else (the API logic, taxonomy, aggregator, snapshot/async model, gotcha
handling) is portable as-is. The modules import each other by bare name (e.g.
import market_taxonomy) — keep them in the same package or adjust imports.
*_signals.py.CORE_MODULES,
REFERENCE_MODULES, keyword lists, and MODULE_CAPS for your audience. The default
folds sports/world-cup/crypto into a collapsed "reference" group.Read reference/apis.md and reference/architecture.md next.
📦 https://github.com/simonlin1212/globalpercent — Star ⭐ 是最好的支持
全球宏观预期概率面板 —— 2 数据源 · 5 端点 · 10 模块 · 零鉴权零账户
一个自包含的 AI Skill（搭建指南 + 参考代码），教你的 AI 编程助手把 Polymarket + Kalshi 两个公开预期概率源的数据，整合成投研系统里的一块「全球宏观预期概率面板」：自动分类成宏观模块（货币政策 / 宏观经济 / 地缘 / 政治选举 / 股指大宗 / AI 科技…），一眼看清全市场的预期概率状态，作为情绪 / 风险温度计叠加层。
它是温度计，不是交易信号。 一份合约报价 0.62 = 市场用真金白银押注该事件有 62% 的预期概率发生。读这些数字完全免费、无需账户、无需钱包——只有「下场交易」才需要账户。所以你能零成本拿到一份全球、资金背书的宏观情绪读数（Fed 决议、地缘、AI 里程碑…），用来读「市场情绪/周期位置/催化剂时点」，而不是用来决定买什么。
这不是一次 API 调用就完事。价值在于系统：统一 schema、分类、去重、多源合并、快照缓存、异步刷新、标题翻译，以及十几个每个都要花真实调试时间才能踩明白的坑——全部封装在这份 Skill 里，你不用重新发现一遍。
兼容 Claude Code · Codex · OpenClaw
Skill 文件本质是结构化 Markdown + 内嵌 Python/React，任何支持上下文注入的 AI 编程助手都能用。
所有接口已于 2026-06-05 对照线上 API 实测验证（字段名会变——尤其 Kalshi 已改过一次，贴代码前请用一行 curl 复验）。
全球宏观预期概率面板 · 数据管道
│
Polymarket Gamma API ─┐
├─► 每个源的 fetcher 把每条市场整形成同一套 schema
Kalshi events API ────┘ (question, prob_yes, change_24h, volume_24h, source, …)
│
共享 taxonomy 把每条分到一个模块
(货币政策/宏观经济/地缘政治/政治选举/股指大宗/AI科技 + 参考组)
│
聚合器：合并 → 按模块分组 → 限流刷屏 →
翻译标题 → 钉一份磁盘 SNAPSHOT
│
/pulse/overview (快照秒开；刷新 = 后台异步重建)
│
React 面板：模块分区 + 来源徽章 + 趋势图
2 步，给你的 AI 助手喂这份 skill。
# 1. 创建 skill 目录
mkdir -p ~/.claude/skills/globalpercent
# 2. 把整个仓库克隆/下载进去（SKILL.md + code/ + reference/）
git clone https://github.com/simonlin1212/globalpercent.git \
~/.claude/skills/globalpercent
然后对 AI 助手说一句：
「用 globalpercent 这份 skill，把一块全球宏观预期概率面板搭进我的 app。」
它会读 reference/apis.md（端点 + 字段 + 每个坑）和 reference/architecture.md（设计 + 异步刷新 + 空拉不覆盖等铁律），再把 code/ 里的后端 + 前端移植进你的技术栈。
依赖：后端只用
httpx（异步 HTTP）。翻译步骤可选（要么接你自己的 LLM，要么直接删掉走英文）。前端是 React + echarts，可换成你自己的图表库。
Codex / OpenClaw 用户： 把 SKILL.md + reference/ 的内容贴入系统 prompt 或项目上下文，内嵌代码可直接移植。
读预期概率不需要账户；只有「交易」才需要 Key。下列端点 2026-06-05 实测全部可用。
| 端点 | 数据 | 实测状态 |
|---|---|---|
Gamma /markets |
每条市场的 question / outcomes / outcomePrices（=概率） / volume24hr / clobTokenIds / 24h&7d 变动 / slug / 到期日 | ✅ 服务端按 volume24hr 排序，limit≤100 翻页 |
CLOB /prices-history |
单个 outcome token 的概率时间序列 {t, p}（趋势图用） |
✅ interval=1d/1w/1m/max + fidelity |
CLOB /midpoint |
单 token 的实时中间价 | ✅ 返回 {mid} |
⚠️ Polymarket 的
outcomes/outcomePrices/clobTokenIds是 JSON 编码的字符串，不是数组——要json.loads()（见_parse_json_field）。
| 端点 | 数据 | 实测状态 |
|---|---|---|
/events?with_nested_markets=true |
每个 event 带原生 category + 嵌套 markets（含 *_dollars 报价 + 成交量），分页 cursor |
✅ 慢且重（200 event/页，2–16s/页，全量 1–8 分钟→必须异步刷新） |
/markets?series_ticker=KXFED |
按系列过滤的市场（如 Fed），快（~2s） | ✅ series_ticker 可用；category 过滤器失效、无成交量排序 |
⚠️⚠️ Kalshi 字段已在 2026-06 改名为
*_dollars：旧的 cents 字段（last_price/yes_bid/yes_ask）现在返回None。要用yes_ask_dollars/last_price_dollars/previous_price_dollars/volume_24h_fp/liquidity_dollars等。本仓库代码已全部按新字段名实现并实测通过。
分类关键词优先、Kalshi 原生 category 兜底，高信号先判（地缘 在 选举 前、加密 在 股指 前、货币 在 宏观 前），world cup/FIFA 最先判进体育。
| 核心模块（展开显示） | 参考模块（折叠显示） |
|---|---|
| 货币政策 · 宏观经济 · 地缘政治 · 政治选举 · 股指大宗 · AI科技 | 加密 · 体育 · 娱乐 · 其他 |
每个模块按 24h 成交量限 top-N（MODULE_CAPS），防止「LA 市长」「每日油价/天气」这类刷屏事件淹没宏观信号。
跟你的 AI 助手说这些话就能驱动这份 skill：
| 场景 | 说什么 |
|---|---|
| 搭面板 | 「用 globalpercent 把一块宏观预期概率面板搭进我的 dashboard」 |
| 只接 Polymarket | 「只接 Polymarket Gamma，先跑通地缘和 AI 两个模块」 |
| 加 Kalshi 宏观 | 「再接上 Kalshi 的 Fed/CPI/就业，做成异步刷新」 |
| 调分类 | 「帮我把 taxonomy 的关键词改成更偏 A 股投研的模块」 |
| 趋势图 | 「给 Polymarket 行加一条概率趋势折线图」 |
| 纯英文 | 「我不需要中文翻译，删掉 translate 那一步」 |
| 能力 | 说明 |
|---|---|
| 双源合并 | Polymarket（地缘/选举/加密深度）+ Kalshi（宏观经济结构）整形成同一套 schema，「加第三个源」= 加一个 *_signals.py |
| 多腿/标量事件折叠 | 油价/CPI/气温这类被拆成 10–40 个阈值腿的标量事件，折叠到最接近 50% 的那条腿（市场隐含中位水平），并打 pick_label 标签（「53% 概率油价高于 $4.20」），不会出现「油价: 2%」这种无意义数字 |
| 快照模型 | 一次拉取+翻译要数秒到数分钟，所以聚合结果钉到磁盘 JSON，普通访问秒开（无网络、无 LLM、重启不丢），只有刷新按钮才重拉 |
| 异步刷新 | Kalshi 全量 1–8 分钟，刷新立即返回旧快照 + updating:true，后台 asyncio.create_task 重建，前端轮询直到 as_of 推进 |
| 空拉永不覆盖（铁律） | Kalshi 边缘节点会间歇性 HTTP 000/SSLEOFError，每页重试 + 一旦拉到 0 条绝不落盘、返回上一份好快照——「API 短暂挂了」绝不能变成「我们把数据删了」 |
| 翻译限流 | 标题按英文原文缓存（永久），小批量(4)+批间延迟(~0.8s)+多轮重试(连续两轮无进展才停)+每批落盘，绕开推理模型大批量返空 + provider 限流两个坑 |
| venue | 公开 API | 结论 |
|---|---|---|
| Polymarket | ✅ Gamma + CLOB，全公开 | 真金白银，地缘/加密/选举最深 —— 首选 |
| Kalshi | ✅ events/markets，只读无需 Key | 真金白银，美国 CFTC 监管，宏观经济结构最干净 —— 首选 |
| Manifold | ✅ api.manifold.markets/v0/markets |
玩具币 → 信号噪声大，只作补充 |
| Metaculus | 有 API 但 403 Cloudflare 拦服务器/curl |
预测社区（非资金），需浏览器/UA 绕过 |
| PredictIt | 403 Cloudflare + 平台衰退 |
跳过 |
结论：Polymarket + Kalshi 是两个免费、无鉴权、资金背书、值得搭建的源，且互补——Polymarket 管地缘/加密/选举深度，Kalshi 管干净的宏观经济结构。
Q: 这能直接跑起来吗？
这是搭建指南 + 参考代码 Skill（类似 a-stock-data 的形态），不是开箱即用的 app。把 code/ 移植进你自己的 FastAPI + React 栈，有 3 个环境相关的替换点（数据目录 / LLM 客户端 / UI 组件库），SKILL.md 里都标注了。
Q: 为什么是温度计不是交易工具？ 70–84% 的预期概率市场参与者是亏钱的，优势属于做市的高频，不属于「预测得准的 AI」。把面板当作「市场情绪如何」来读，永远不要当买卖信号。
Q: Kalshi 拉取怎么这么慢？ Kalshi 没有服务端成交量排序，热门市场被几千个一日体育/天气合约埋着，所以必须把整个 open event book（~7000 event，~35 页）全拉下来客户端排序。全量 1–8 分钟——这就是刷新必须异步的原因。
Q: 字段名会不会又变？
会。Kalshi 2026-06 刚把价格字段改成 *_dollars，旧字段返回 None。贴代码前用一行 curl 复验字段名（README 顶部端点表给了实测日期）。
Q: 翻译步骤必须要吗？ 不要。受众读英文的话，直接删掉 translate 那一步走纯英文。它是纯辅助层。
Q: 不用 Claude Code 能用吗？ 能。SKILL.md + reference/ 是 Markdown，code/ 是标准 Python/React，任何 AI 编程助手或人工都能读取移植。
见 CHANGELOG.md。
如果这份 skill 帮到了你的投研工作流，欢迎请作者喝杯咖啡 ☕
想接更多预期概率源（Manifold / 自定义）或更多模块？欢迎开 Issue 提需求，赞助者的 Issue 优先处理。
本项目仅提供公开预期概率数据的获取与可视化工具，是一块情绪 / 风险温度计，不构成任何投资建议，也不是交易信号。所有数据来自第三方公开 API，准确性与可用性以来源为准。市场有风险，决策需谨慎。
Apache License 2.0 — 自由使用，注明出处即可。
作者： Simon 林 · 抖音「Simon林」 · 公众号「硅基世纪」
A global-macro-probability panel — 2 sources · 5 endpoints · 10 modules · zero auth, zero account.
A self-contained AI Skill (build guide + reference code) that teaches your AI coding agent to merge public probability data from Polymarket + Kalshi into a single global-macro-probability panel for an investment-research system: it classifies every market into macro modules (monetary policy / macro economy / geopolitics / elections / indices & commodities / AI…) and surfaces the whole market's expected-probability state at a glance as a sentiment / risk overlay.
It's a thermometer, not a trading signal. A contract trading at $0.62 means the market bets $0.62 = 62% probability the event happens. Reading those numbers is free, no account, no wallet — only trading needs one. So you get a global, money-backed read on macro mood (Fed, geopolitics, AI milestones…) at zero cost — to read regime/cycle/catalyst-timing, not to pick what to buy.
It is NOT a single API call. The value is the system: one common schema, classification, dedup, multi-source merge, snapshot caching, async refresh, translation, and a dozen non-obvious gotchas — all captured so you don't rediscover them.
Compatible with Claude Code · Codex · OpenClaw.
All endpoints were verified against the live APIs on 2026-06-05 (field names change — Kalshi already renamed once; re-verify with a quick curl before trusting any field).
Polymarket Gamma API ─┐
├─► per-source fetchers shape every market into ONE common schema
Kalshi events API ────┘ (question, prob_yes, change_24h, volume_24h, source, …)
│
shared taxonomy classifies each into a module
(monetary/macro/geo/elections/indices/AI + reference group)
│
aggregator: merge → group by module → cap floods →
translate titles → pin a disk SNAPSHOT
│
/pulse/overview (instant from snapshot; refresh = async rebuild)
│
React panel: module sections + source badges + trend chart
mkdir -p ~/.claude/skills/globalpercent
git clone https://github.com/simonlin1212/globalpercent.git \
~/.claude/skills/globalpercent
Then tell your agent: "Use the globalpercent skill to build a macro-probability panel into my app." It reads reference/apis.md + reference/architecture.md, then ports code/ into your stack (3 documented env-specific swap points: data dir / LLM client / UI components). Backend needs only httpx; translation is optional; frontend is React + echarts.
Polymarket — Gamma /markets (question / outcomes / outcomePrices = probability / volume24hr / clobTokenIds / 24h&7d change / slug), CLOB /prices-history ({t,p} series for the trend chart), CLOB /midpoint. ⚠️ outcomes/`out
globalpercent is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by simonlin1212. 全球宏观预期概率面板 (AI Skill) — Polymarket+Kalshi 双源 · 5端点 · 10模块 · 零鉴权 | Global-macro-probability panel skill for AI coding assistants (prediction-market data pipeline). It has 50 GitHub stars.
globalpercent's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/simonlin1212/globalpercent" and add it to your Claude Code skills directory (see the Installation section above). globalpercent ships a SKILL.md manifest, so compatible agents can discover and load it automatically.
globalpercent is primarily written in Python. It is open-source under simonlin1212 on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh globalpercent against similar tools.
No comments yet. Be the first to share your thoughts!