Search Polymarket for prediction markets matching a topic, open the most relevant one, and return Zod-validated structured metadata: question, current YES/NO implied probabilities, USD volume (cumulative + rolling windows), resolution date, and top traders (largest holders per outcome).
Works with
AI-first code editor with Composer
Before installing skills in Cursor, ensure your development environment meets these requirements:
node --versionpolymarket-researchExecute the skills CLI command in your project's root directory to begin installation:
Fetches polymarket-research from polymarket.com/polymarket-research-0rhj5o and configures it for Cursor.
The CLI shows a list of agents. Use arrow keys and space to select Cursor:
Confirm successful installation by checking the skill directory location:
Restart Cursor to activate polymarket-research. Access via /polymarket-research in your agent's command palette.
We perform automated surface-level scans (Gen AI Scanner, Socket, Snyk) during installation. These checks detect common vulnerabilities but do not guarantee complete security. Always review skill source code and verify the publisher's reputation before production use.
Skills execute code in your environment. Always review source, verify the publisher, and test in isolation before production.
Submit your Claude Code skill and start earning
Automate repetitive workflows and reduce manual effort
Example
Generate reports, summarize documents, draft communications
Save 3-5 hours per week on routine tasks
Learn new skills, understand complex topics, get expert guidance
Example
Explain concepts, provide examples, suggest learning resources
Accelerate learning and skill development by 2x
Enhance output quality through reviews, suggestions, and refinements
Example
Review drafts, suggest improvements, catch errors
Improve work quality by 30-40% with less effort
0
total installs
0
this week
0
upvotes
Run in your terminal
0
installs
0
this week
—
stars
| name | polymarket-research |
| title | Polymarket Topic Research |
| description | >- Search Polymarket for prediction markets matching a topic, open the most relevant one, and return Zod-validated structured metadata: question, current YES/NO implied probabilities, USD volume (cumulative + rolling windows), resolution date, and top traders (largest holders per outcome). |
| website | polymarket.com |
| category | prediction-markets |
| tags | - polymarket - prediction-markets - odds - research - read-only - api |
| source | 'browserbase: agent-runtime 2026-05-21' |
| updated | '2026-05-21' |
| recommended_method | api |
| alternative_methods | - method: browser rationale: >- Polymarket's public Gamma + Data REST APIs (gamma-api.polymarket.com, data-api.polymarket.com) are unauthenticated, CORS-open, and serve the exact data the polymarket.com UI consumes. Scraping the /event/{slug} page works only as a degraded fallback — holder amounts arrive as abbreviated strings ('11.2K') in the HTML and require clicking the Top Holders tab to even render. Use the API. |
| verified | false |
| proxies | false |
Given a free-text topic (e.g. "trump", "bitcoin $150k", "Iran nuclear deal"), search Polymarket for matching prediction markets, pick the most relevant event, and return a Zod-validated structured payload containing the market question, current YES/NO odds (implied probability + last trade price + best bid/ask), volume, resolution / end date, and the top traders (largest holders) on each side. Read-only — never places an order, never signs a wallet transaction, never interacts with the order book.
polymarket.com/event/... HTML — the public Gamma + Data APIs return everything the UI shows, in JSON, with no auth.Polymarket exposes a fully public, unauthenticated, CORS-open REST surface that the polymarket.com front-end itself consumes. Three hosts cover this task end-to-end:
| Host | Purpose |
|---|---|
gamma-api.polymarket.com | Search, event metadata, market metadata, prices |
data-api.polymarket.com | Top holders (per-outcome leaderboard), recent trades, positions |
clob.polymarket.com | Order book / live spread (only needed if you want raw L2 depth) |
No API key, no JWT, no cookies, no Referer games, no residential proxy — a vanilla fetch from any IP works. Lead with the API. Browser scraping is a last-resort fallback that costs ~30× more turns for strictly less data (holder amounts, event metadata blob, and the AI-generated "context_description" are exposed only on the JSON side).
GET https://gamma-api.polymarket.com/public-search
?q={url-encoded query}
&limit_per_type={N}
&events_status=active # optional: 'active' | 'resolved' | omit for both
Response shape:
{
"events": [ { "id", "slug", "title", "description", "endDate", "volume",
"liquidity", "openInterest", "commentCount", "tags": [...],
"markets": [ {...market objects...} ],
"eventMetadata": { "context_description": "AI-generated..." } } ],
"tags": [],
"profiles": [],
"pagination": {...}
}
Each event may contain one market (binary Yes/No question — markets.length === 1) or many markets (categorical / multi-outcome event — each child market is a Yes/No on a single option, with the human-readable option in groupItemTitle, e.g. "Japan / Korea", "Friend of mine").
public-search already returns events ordered by an internal relevance score. The first non-archived, non-closed event in events[] is the default "most relevant". Optionally re-rank by combining score with volume if you want to bias toward liquid markets:
const top = events
.filter(e => !e.closed && !e.archived)
.sort((a, b) => (b.volume ?? 0) - (a.volume ?? 0))[0]; // or keep API order
Pull the chosen event's markets[]. For each market:
Field on market object | Meaning |
|---|---|
question | The literal Yes/No question (string) |
outcomes | JSON-string array, e.g. '["Yes", "No"]' — JSON.parse it |
outcomePrices | JSON-string array of stringified floats '["0.9995", "0.0005"]' — JSON.parse then parseFloat. These are the implied probabilities (0..1), not USD prices |
lastTradePrice | Float, last trade midprice (probability) |
bestBid / bestAsk | Floats, current order-book quote |
spread | bestAsk - bestBid |
volumeNum / volume | Cumulative USD volume (number; the string volume is the same value) |
volume24hr / volume1wk / volume1mo / volume1yr | Rolling-window USD volumes |
liquidityNum / liquidityClob | Current CLOB liquidity (USD) |
endDate / endDateIso | ISO-8601 resolution date |
conditionId | 0x… 32-byte hex — primary key for the Data API |
clobTokenIds | JSON-string array of two large decimal token IDs — clobTokenIds[0] = YES token, clobTokenIds[1] = NO token |
closed / active / archived | Lifecycle flags |
groupItemTitle | Set when the event is multi-outcome; the human label for this option |
umaResolutionStatuses | JSON-string array; contains "proposed" / "disputed" during resolution windows |
description | Full resolution-rules text |
The event-level eventMetadata.context_description is an AI-generated trader-context blurb updated every few hours — useful as a "why is this market moving" summary.
GET https://data-api.polymarket.com/holders
?market={conditionId} # the 0x… hex from step 3
&limit={N} # default returns top ~10 per outcome
The market query param is the conditionId, not the integer market id and not a CLOB token id. The endpoint will 400 if the param is missing.
Response shape — array of one entry per outcome token (Yes/No → 2 entries; multi-outcome neg-risk markets → N entries):
[
{ "token": "40323545806742...",
"holders": [
{ "proxyWallet": "0x2785…",
"name": "poorsob", // public display name; may be empty
"pseudonym": "Oddball-Maniac", // auto-generated fallback; always set
"amount": 2989847.91, // shares held (numeric)
"outcomeIndex": 0, // 0 → Yes, 1 → No (binary markets)
"verified": false,
"displayUsernamePublic": true,
"bio": "", "profileImage": "" }
] },
{ "token": "5845206889...", "holders": [ … outcomeIndex: 1 → No side … ] }
]
outcomeIndex maps positionally to the outcomes / outcomePrices arrays from the Gamma payload, so token 0 = Yes, token 1 = No on standard binary markets. To get a wallet's profile URL: https://polymarket.com/profile/{proxyWallet}.
Optional: &sort=value ranks by USD value instead of share count (default sorts by raw amount).
GET https://data-api.polymarket.com/trades
?market={conditionId}
&limit={N}
Returns per-trade rows with proxyWallet, side (BUY/SELL), outcome ("Yes"/"No"), size, price, timestamp (Unix seconds), and the trader's name/pseudonym. Useful when "top traders" should mean "most recently active" rather than "largest current holders".
See ## Expected Output for the canonical Zod schema. Key transforms vs. raw API:
JSON.parse both outcomes and outcomePrices (they ship as JSON-encoded strings inside a JSON response — yes, double-encoded).volume, liquidity) to floats; the API serves both volume (string) and volumeNum (number) — prefer volumeNum.holder.name || holder.pseudonym.marketUrl = https://polymarket.com/event/{event.slug} (event page, not per-child-market — Polymarket has no per-market URL for multi-outcome events; the /event/... slug is the canonical permalink).If gamma-api.polymarket.com ever returns a non-2xx, fall back to scraping. A stealth + residential-proxy session is mandatory only because the marketing site sits behind Cloudflare; the API hosts themselves do not require stealth.
sid=$(browse cloud sessions create --keep-alive --verified --proxies \
| node -e "let s='';process.stdin.on('data',c=>s+=c).on('end',()=>process.stdout.write(JSON.parse(s).id))")
export BROWSE_SESSION="$sid"
# Search: /markets?_q={topic} redirects to /predictions/{slug} when the topic is a tag,
# OR returns a generic list page. Prefer the API.
browse open "https://polymarket.com/markets?_q=trump" --remote --session "$sid"
browse wait load --remote --session "$sid"
browse wait timeout 3000 --remote --session "$sid"
browse get markdown body --remote --session "$sid" # parse cards from markdown
# Detail page (must visit /event/{slug} — not /market/...)
browse open "https://polymarket.com/event/trump-kiss-by-may-31" --remote --session "$sid"
browse wait load --remote --session "$sid"
browse wait timeout 4000 --remote --session "$sid"
browse snapshot --remote --session "$sid"
# Click the "Top Holders" tab to render the holder list in the DOM:
browse click <Top Holders heading ref> --remote --session "$sid"
browse wait timeout 2500 --remote --session "$sid"
browse get markdown body --remote --session "$sid"
The HTML carries question / volume / end date / odds and (after clicking the "Top Holders" tab) the holder list — but extracting holder amounts cleanly from markdown is brittle (numbers are abbreviated to "11.2K", "6.0K", etc.). Use the API for the structured numbers; only use the browser to confirm the page exists.
outcomes and outcomePrices are double-encoded JSON. They arrive as strings like '["Yes", "No"]' inside the JSON response. Forgetting to JSON.parse them silently yields strings where you expected arrays — the most common bug when consuming this API for the first time.outcomePrices are implied probabilities (0..1), not USD prices. 0.9995 means "the market thinks YES is 99.95% likely", not "$0.9995 per share" (though Polymarket's UI conveniently lets you read them either way — shares pay $1 on resolution, so probability == price). Multiply by 100 if your downstream consumer wants percent.market param on /holders is the conditionId (0x… hex), NOT the integer id and NOT the clobTokenIds decimal. Passing the wrong one returns 400 {"error":"required query param 'market' not provided"} (misleading — the param IS present, just unrecognized).public-search never returns "no results". When the query has no good match, the endpoint returns the top globally-ranked events as a fallback (e.g. searching xyzzy_no_such_thing returned a Golden Globes event). To detect "no real hits", check whether events[0].title contains the query tokens (case-insensitive), or filter by volume > threshold + a fuzzy-match score. Don't treat a 200 + non-empty events[] as proof the query was relevant./markets (Gamma) is deprecated, Sunset: Fri, 01 May 2026. Use /markets/keyset for paginated listing or /public-search / /events for everything else. The deprecated endpoint still responds with 200 + Warning: 299 - "use /markets/keyset" and Deprecation: true headers — don't ignore them in long-lived code.conditionId and groupItemTitle. The "current odds" for option "Japan / Korea" is outcomePrices[0] (Yes) of that child market, not a column in a parent market. negRisk: true is a hint that the children are mutually exclusive (only one can resolve Yes), but negRisk: false multi-market events also exist (independent Yes/No on each option). Schema accordingly.conditionId survives event mutation; id and slug do not. Polymarket occasionally re-slugs events (typo fixes, re-launches). Cache markets by conditionId if you persist anything across sessions./event/{slug} is the canonical URL — /market/{...} does not exist on the public site. For multi-outcome events, every child market shares the same /event/{slug} page; there's no per-child-market URL.outcomeIndex is 0=Yes/1=No on binary markets, but is 0..N-1 on multi-outcome NEG-risk events — there's no "neg-risk" flag in the holder response itself. Read outcomeIndex against the parent market's outcomes array to get the human label.&sort=value), (b) most recent trades (/trades?market=...), (c) all-time PnL leaders (no public endpoint — would require summing trades over history). Document which you used.pseudonym is always set; name is often empty. Some traders haven't set a display name; their wallet shows as Oddball-Maniac-style auto-generated handles. Use name || pseudonym as the resolved display label.umaResolutionStatuses signals an in-progress resolution. A non-empty array like ["proposed"] or ["disputed"] means UMA (the optimistic oracle) is mid-resolving — the outcomePrices may show 0.9995 / 0.0005 long before the market is actually closed: true. If your downstream consumer needs "is the answer final?", check closed === true, not just whether prices are pegged.gamma-api and data-api; cache hits return in ~50ms and bypass any origin throttle. Don't blast — be polite./predictions/{topic} is a marketing-SEO page, not search. It returns hand-curated events tagged with that topic (e.g. /predictions/trump), not full-text matches. The browser-fallback's /markets?_q={query} URL silently rewrites to /predictions/{slug} when {slug} matches a known tag — the search experience is "tag-first, full-text fallback". The Gamma API's public-search is the only full-text path.The skill returns one of three shapes. Zod schema:
import { z } from "zod";
const Trader = z.object({
proxyWallet: z.string().regex(/^0x[0-9a-fA-F]{40}$/),
displayName: z.string(), // name || pseudonym
pseudonym: z.string(),
amount: z.number().nonnegative(), // shares held
outcomeIndex: z.number().int().min(0), // 0=Yes, 1=No (binary)
outcomeLabel: z.string(), // "Yes" / "No" / "Japan / Korea" / ...
verified: z.boolean(),
profileUrl: z.string().url(), // https://polymarket.com/profile/{proxyWallet}
});
const Market = z.object({
conditionId: z.string().regex(/^0x[0-9a-fA-F]{64}$/),
question: z.string(),
groupItemTitle: z.string().nullable(), // populated for multi-outcome children
outcomes: z.array(z.string()), // e.g. ["Yes","No"]
prices: z.object({ // implied probabilities, 0..1
yes: z.number().min(0).max(1),
no: z.number().min(0).max(1),
lastTrade: z.number().min(0).max(1).nullable(),
bestBid: z.number().min(0).max(1).nullable(),
bestAsk: z.number().min(0).max(1).nullable(),
spread: z.number().min(0).max(1).nullable(),
}),
volume: z.object({
total: z.number().nonnegative(),
day: z.number().nonnegative().nullable(),
week: z.number().nonnegative().nullable(),
month: z.number().nonnegative().nullable(),
year: z.number().nonnegative().nullable(),
}),
liquidity: z.number().nonnegative().nullable(),
endDate: z.string().datetime(), // ISO-8601
closed: z.boolean(),
active: z.boolean(),
topTraders: z.array(Trader), // typically top 5-10 per side, concatenated
});
const ResearchResult = z.discriminatedUnion("status", [
// 1. Found a clearly relevant binary market
z.object({
status: z.literal("ok"),
query: z.string(),
event: z.object({
id: z.string(),
slug: z.string(),
title: z.string(),
url: z.string().url(), // https://polymarket.com/event/{slug}
description: z.string(),
contextDescription: z.string().nullable(), // eventMetadata.context_description
tags: z.array(z.object({ id: z.string(), label: z.string(), slug: z.string() })),
volume: z.number().nonnegative(),
liquidity: z.number().nonnegative(),
openInterest: z.number().nonnegative().nullable(),
}),
markets: z.array(Market).min(1), // 1 entry for binary, N for multi-outcome
multiOutcome: z.boolean(), // true iff markets.length > 1
fetchedAt: z.string().datetime(),
}),
// 2. Search returned events but none look relevant (top result doesn't contain query tokens)
z.object({
status: z.literal("no_relevant_match"),
query: z.string(),
topGuess: z.object({ slug: z.string(), title: z.string(), url: z.string().url() }),
reason: z.string(), // e.g. "top result title does not contain any query token"
}),
// 3. Hard failure (API non-2xx after retries)
z.object({
status: z.literal("error"),
query: z.string(),
httpStatus: z.number().int(),
message: z.string(),
}),
]);
Example status: "ok" payload (binary market):
{
"status": "ok",
"query": "trump kiss",
"event": {
"id": "59112",
"slug": "trump-kiss-by-may-31",
"title": "Trump kiss by May 31?",
"url": "https://polymarket.com/event/trump-kiss-by-may-31",
"description": "This market will resolve to \"Yes\" if Donald Trump and any other person kiss by the specified date…",
"contextDescription": "A recent viral video capturing Donald Trump exchanging a kiss…",
"tags": [{ "id": "2", "label": "Politics", "slug": "politics" }],
"volume": 6310760.45,
"liquidity": 8945123.0,
"openInterest": null
},
"markets": [
{
"conditionId": "0xcef981d46a1039b6ae02f578d2208302a8a3d63465d24363a1d65a86835a1ae8",
"question": "Trump kiss by May 31?",
"groupItemTitle": null,
"outcomes": ["Yes", "No"],
"prices": {
"yes": 0.9995, "no": 0.0005,
"lastTrade": 0.9995, "bestBid": 0.999, "bestAsk": 1.0, "spread": 0.001
},
"volume": { "total": 6310760.45, "day": 532019.1, "week": 6310760.45, "month": 6310760.45, "year": 6310760.45 },
"liquidity": 8945123.0,
"endDate": "2026-05-31T00:00:00Z",
"closed": false,
"active": true,
"topTraders": [
{ "proxyWallet": "0x2785e7022dc20757108204b13c08cea8613b70ae",
"displayName": "poorsob", "pseudonym": "Oddball-Maniac",
"amount": 2989847.91, "outcomeIndex": 0, "outcomeLabel": "Yes",
"verified": false, "profileUrl": "https://polymarket.com/profile/0x2785e7022dc20757108204b13c08cea8613b70ae" },
{ "proxyWallet": "0x558ee6...", "displayName": "perseusplus", "pseudonym": "...",
"amount": 4044588.47, "outcomeIndex": 1, "outcomeLabel": "No",
"verified": false, "profileUrl": "https://polymarket.com/profile/0x558ee6..." }
]
}
],
"multiOutcome": false,
"fetchedAt": "2026-05-21T00:08:00Z"
}
Example status: "no_relevant_match" payload:
{
"status": "no_relevant_match",
"query": "xyzzy_no_such_thing",
"topGuess": {
"slug": "golden-globes-best-motion-picture-musical-or-comedy-winner",
"title": "Golden Globes: Best Motion Picture – Musical or Comedy Winner",
"url": "https://polymarket.com/event/golden-globes-best-motion-picture-musical-or-comedy-winner"
},
"reason": "top result title does not contain any query token; Polymarket public-search returns a fallback ranking when no good match exists"
}
Prerequisites
Time Estimate
15-45 minutes depending on use case complexity
Steps
Common Pitfalls
✓ Do
✗ Don't
💡 Pro Tips
✓ Use when
Use when skill capabilities match your task, clear ROI on time saved, and you can validate outputs. Best for repetitive tasks, learning, and quality improvement.
✗ Avoid when
Avoid when task requires deep expertise you can't validate, involves sensitive decisions, or when learning process is more valuable than speed of completion.
BuilderIO/skills
ailabs-393/ai-labs-claude-skills
affaan-m/everything-claude-code
kostja94/marketing-skills
aj-geddes/useful-ai-prompts
davila7/claude-code-templates
polymarket-research has been reliable in day-to-day use. Documentation quality is above average for community skills.
Useful defaults in polymarket-research — fewer surprises than typical one-off scripts, and it plays nicely with `npx skills` flows.
polymarket-research has been reliable in day-to-day use. Documentation quality is above average for community skills.
Solid pick for teams standardizing on skills: polymarket-research is focused, and the summary matches what you get after install.
We added polymarket-research from the explainx registry; install was straightforward and the SKILL.md answered most questions upfront.
Solid pick for teams standardizing on skills: polymarket-research is focused, and the summary matches what you get after install.
We added polymarket-research from the explainx registry; install was straightforward and the SKILL.md answered most questions upfront.
We added polymarket-research from the explainx registry; install was straightforward and the SKILL.md answered most questions upfront.
Solid pick for teams standardizing on skills: polymarket-research is focused, and the summary matches what you get after install.
polymarket-research reduced setup friction for our internal harness; good balance of opinion and flexibility.
showing 1-10 of 50