Monitor Prediction Markets

${var} — Platform selector with an optional single-market override:

• empty ("") — run both platforms from their watchlists.
• polymarket — run Polymarket's whole watchlist (skills/monitor-polymarket/watchlist-polymarket.md).
• kalshi — run Kalshi's whole watchlist (skills/monitor-polymarket/watchlist-kalshi.md).
• polymarket:<event-slug> — one ad-hoc Polymarket event (e.g. polymarket:us-x-iran-ceasefire-by).
• kalshi:<event-ticker> — one ad-hoc Kalshi event (e.g. kalshi:KXGDP-26Q2).

Read memory/MEMORY.md for context. Read the last 2 days of memory/logs/ to compare against previous readings and flag new movers (not repeats of yesterday's news).

Why this skill exists

A table of prices isn't useful. An operator reading this notification wants to answer: "is there a market worth forming a view on right now, and why?" Every rule below exists to push output toward that question — suppress noise, rank by decision value, and demand one line of reasoning per alert. This skill covers two venues — Polymarket (crypto-native, CLOB + comments) and Kalshi (regulated, liquidity-weighted signals) — and dispatches to the branch(es) the selector asks for.

Dispatch

Parse ${var} into a platform choice and an optional single-market override, then run the matching branch(es).

PLATFORM="both"   # both | polymarket | kalshi
SINGLE=""         # optional single event slug (Polymarket) or ticker (Kalshi)

case "${var}" in
  "")            PLATFORM="both" ;;
  polymarket)   PLATFORM="polymarket" ;;
  kalshi)       PLATFORM="kalshi" ;;
  polymarket:*) PLATFORM="polymarket"; SINGLE="${var#polymarket:}" ;;
  kalshi:*)     PLATFORM="kalshi";     SINGLE="${var#kalshi:}" ;;
  *)            # unrecognised prefix — don't guess a venue; fall back to both watchlists
                PLATFORM="both"; SINGLE=""
                echo "unrecognised selector '${var}' — running both watchlists" ;;
esac

• PLATFORM=both → run the Polymarket branch and the Kalshi branch, each from its own watchlist, then emit a combined notification.
• PLATFORM=polymarket → run only the Polymarket branch (whole watchlist, or SINGLE if set).
• PLATFORM=kalshi → run only the Kalshi branch (whole watchlist, or SINGLE if set).

Each branch below is independently executable — skip the one(s) the selector didn't ask for.

Polymarket branch

Data source: Polymarket gamma-api (events, comments) + clob (price history). All endpoints are public — no auth.

Watchlist: skills/monitor-polymarket/watchlist-polymarket.md. Each line is an event slug; add or remove slugs to change what's monitored.

P1. Load watchlist

if [ -n "$SINGLE" ]; then
  SLUGS="$SINGLE"
else
  # One slug per line, skip comments and blanks
  SLUGS=$(grep -v '^#' skills/monitor-polymarket/watchlist-polymarket.md | grep -v '^$')
fi

If the watchlist is empty and no single slug was given, there's nothing to do on Polymarket — note it and move on (don't fabricate a report).

P2. For each event, fetch markets and price history

For each event slug in $SLUGS:

a) Get the event and its markets:

curl -s "https://gamma-api.polymarket.com/events?slug=$SLUG&limit=1"

The response contains the event id, title, and a markets array. Each market has:

• id, question, slug, closed
• outcomePrices — JSON array, index 0 = YES price (0.0–1.0)
• volume24hr, volumeNum, liquidityNum
• clobTokenIds — JSON array, index 0 = YES token, index 1 = NO token

Skip closed markets — they've already resolved.

b) Get 24h price history for each open market:

# YES token is index 0 of clobTokenIds
TOKEN_ID=$(echo "$CLOB_TOKEN_IDS" | python3 -c "import json,sys; print(json.loads(sys.stdin.read())[0])")
curl -s "https://clob.polymarket.com/prices-history?market=$TOKEN_ID&interval=1d&fidelity=60"

Response: { "history": [{ "t": unix_timestamp, "p": "price_string" }, ...] }

c) Calculate 24h stats for each market:

• Open / Close — first and last price in the history
• Change — close minus open, in percentage points (e.g. +4.0pp)
• High / Low — intraday range
• Volume — volume24hr from the market data
• Direction — classify as: surging (>+5pp), rising (+2 to +5pp), stable (−2 to +2pp), falling (−5 to −2pp), crashing (<−5pp)

P3. Fetch comments

For each event, get top comments and latest comments:

EVENT_ID=... # from step P2a

# Top comments by reactions
curl -s "https://gamma-api.polymarket.com/comments?parent_entity_type=Event&parent_entity_id=$EVENT_ID&limit=10&order=reactionCount&ascending=false"

# Latest comments (last 24h chatter)
curl -s "https://gamma-api.polymarket.com/comments?parent_entity_type=Event&parent_entity_id=$EVENT_ID&limit=10&order=createdAt&ascending=false"

Important: parent_entity_type must be Event (capital E).

Each comment has: body, profile.username (often null → use "anon"), reactionCount, createdAt.

From the combined results, pick the 3 most interesting comments per event:

• New comments from the last 24h get priority (they react to recent moves)
• High-reaction comments that are still relevant
• Contrarian takes, insider-sounding analysis, whale callouts, humor

P4. Build the Polymarket report

For each event, produce a summary block:

**[Event Title]** (event_id: N)

| Market | YES | 24h Chg | High/Low | 24h Vol |
|--------|-----|---------|----------|---------|
| [question] | XX.X% | +X.Xpp ▲ | XX–XX% | $X.Xm |
| [question] | XX.X% | -X.Xpp ▼ | XX–XX% | $X.Xm |
...

Biggest mover: "[question]" — [direction] from X% to Y%

Comments:
- [user/anon]: "[comment excerpt]" (X upvotes)
- [user/anon]: "[comment excerpt]"
- [user/anon]: "[comment excerpt]"

Flag any market that moved more than 5 percentage points in 24h — these are the ones worth paying attention to.

P5. Polymarket Network note

curl works — there is no network sandbox. Use WebFetch as a fallback for a flaky public GET:

• WebFetch("https://gamma-api.polymarket.com/events?slug=SLUG&limit=1")
• WebFetch("https://clob.polymarket.com/prices-history?market=TOKEN_ID&interval=1d&fidelity=60")
• WebFetch("https://gamma-api.polymarket.com/comments?parent_entity_type=Event&parent_entity_id=EVENT_ID&limit=10&order=reactionCount&ascending=false")
• All Polymarket gamma-api / clob endpoints are public and need no auth headers.

Kalshi branch

Data source: Kalshi trade-api v2 at https://api.elections.kalshi.com/trade-api/v2. Despite the "elections" subdomain, this provides access to ALL Kalshi markets (economics, climate, tech, politics, etc.). All endpoints are public — no auth required.

Watchlist: skills/monitor-polymarket/watchlist-kalshi.md. Each line is an event ticker; add or remove tickers to change what's monitored.

K1. Load watchlist

if [ -n "$SINGLE" ]; then
  TICKERS="$SINGLE"
else
  TICKERS=$(grep -v '^#' skills/monitor-polymarket/watchlist-kalshi.md | grep -v '^$')
fi

If the watchlist is empty and no single ticker was given, emit MONITOR_KALSHI_NO_CONFIG, notify with a one-line setup hint, and discover trending events for this run only:

curl -s "https://api.elections.kalshi.com/trade-api/v2/events?status=open&with_nested_markets=true&limit=10"

Pick the 5 highest-volume events and monitor those.

K2. For each event, fetch markets, prices, and liquidity

For each event ticker:

a) Event + markets:

curl -s "https://api.elections.kalshi.com/trade-api/v2/events/$EVENT_TICKER?with_nested_markets=true"

Fields used: event_ticker, title, category, mutually_exclusive, markets[] with ticker, title, subtitle, status, yes_bid, yes_ask, last_price, volume, volume_24h, open_interest, close_time, series_ticker.

Skip non-open markets (closed/settled are historical).

b) 24h candlesticks (batch where possible): Prefer the batch endpoint — one call per event, not per market:

END_TS=$(date -u +%s)
START_TS=$((END_TS - 86400))
# Batch: up to 10,000 candlesticks total across requested tickers
curl -s "https://api.elections.kalshi.com/trade-api/v2/markets/candlesticks?tickers=$COMMA_SEP_MARKET_TICKERS&start_ts=$START_TS&end_ts=$END_TS&period_interval=60"

If the batch endpoint errors, fall back to the per-market endpoint:

curl -s "https://api.elections.kalshi.com/trade-api/v2/series/$SERIES_TICKER/markets/$MARKET_TICKER/candlesticks?start_ts=$START_TS&end_ts=$END_TS&period_interval=60"

If both fail for a market, mark its source as SRC=price_only and use last_price vs yesterday's log entry.

c) Orderbook depth (liquidity / conviction signal):

curl -s "https://api.elections.kalshi.com/trade-api/v2/markets/$MARKET_TICKER/orderbook?depth=10"

From the orderbook, compute:

• spread_pp = yes_ask − yes_bid in percentage points. Wide spread = low conviction, thin book.
• depth_usd = sum over top-10 bid levels of price × size (approximation, both sides). This scales how much weight to give a price.

If orderbook fails, mark SRC=no_book and skip the conviction column for that market.

K3. Compute per-market signals

For each open market:

• implied_prob = last_price as a percentage (0.62 → 62%). Report this, not cents.
• chg_pp = close − open from candlesticks, in percentage points.
• high / low = intraday range.
• vol_24h_usd ≈ volume_24h × avg(open, close) (Kalshi reports contract count — convert so readers can compare across markets).
• spread_pp and depth_usd from step K2c.
• move_score = |chg_pp| × log10(max(vol_24h_usd, 100)). This is the key ranking signal — a 3pp move on a $200k market outranks a 5pp move on a $5k market. It prevents thin-book noise from dominating.

Direction label (from chg_pp): surging (>+5pp), rising (+2 to +5), stable (−2 to +2), falling (−5 to −2), crashing (<−5).

Conviction label (from spread_pp): tight (<2pp), loose (2–5pp), thin (>5pp, treat price skeptically).

K4. Decide what's worth saying — suppression rules

Before building the report, drop markets that fail ALL of these gates:

• |chg_pp| >= 2 AND vol_24h_usd >= $1,000, OR
• vol_24h_usd >= $25,000 (large volume alone is signal even if price didn't move much), OR
• open_interest grew >30% vs yesterday's log, if yesterday's log has the data.

If a market appeared in yesterday's log with the same direction and a chg_pp within ±1pp of today's, treat it as "continued from yesterday" and demote it — mention once at the event level, don't re-alert.

Hard alert threshold: |chg_pp| >= 5 AND conviction != thin. These go to the ALERTS block and require a "why it matters" line.

K5. Global ranking

Rank events by the max move_score of any market within them. Cap the report at the top 5 events. Markets within an event are listed in descending move_score order, capped at 3 per event (mention "+N more" if truncated).

K6. Build the Kalshi report

*Kalshi monitor — ${today}*
verdict: [1 sentence — dominant theme or "all quiet"]

**[Event Title]** (EVENT_TICKER) — category
| Market | prob | Δ24h | range | vol | spread |
|--------|------|------|-------|-----|--------|
| [title] | 62% | +4.1pp ▲ | 56–65% | $82k | 1pp |
| [title] | 23% | −2.8pp ▼ | 22–28% | $14k | 3pp |
mover: [title] — rising on $82k vol, tight book

[next event ...]

**ALERTS** (moved >5pp on non-thin book)
- [event/market]: 34% → 51% — *why it matters:* [one sentence grounded in the move's volume, spread, or news context if obvious from titles]
- ...

**Trending (not tracked)**
- [event] — $Xk 24h vol, consider adding
- ...

sources: events=ok candlesticks=ok|degraded|fail orderbook=ok|degraded|fail

Rules for the verdict line:

• If no alerts AND no market moved >2pp: say "all quiet — [N] events, [M] markets tracked, no moves worth flagging".
• If one theme dominates (most big moves in one category): name it. E.g. "GDP markets repriced down after Q1 print; inflation markets unchanged".
• Never hedge. If you're not sure, say "mixed signals" and stop.

Rules for "why it matters":

• Must reference at least one of: volume (is this real money?), spread (is this consensus?), prior log state (is this new?), or a plausible news trigger inferable from the market title.
• Max 15 words. No filler like "interesting move" or "worth watching".

K7. Discover notable trends

curl -s "https://api.elections.kalshi.com/trade-api/v2/events?status=open&with_nested_markets=true&limit=50"

Scan for events with high volume_24h (top 10) whose tickers are not in the watchlist. Mention 1–2 in the "Trending (not tracked)" block, only if their 24h volume exceeds the median volume of tracked events.

K8. Kalshi status codes (end-of-run)

• MONITOR_KALSHI_OK — ran fully, had data, at least one event processed.
• MONITOR_KALSHI_DEGRADED — partial data (some markets fell back to price_only or no_book); report still sent.
• MONITOR_KALSHI_NO_CONFIG — empty watchlist and no single ticker; discovered trending events and notified with setup hint.
• MONITOR_KALSHI_ERROR — events endpoint failed entirely or zero markets resolved; notify with the failure, don't fake a report.

K9. Kalshi Network note

curl works — there is no network sandbox. Use WebFetch as a fallback for a flaky public GET:

• WebFetch("https://api.elections.kalshi.com/trade-api/v2/events/EVENT_TICKER?with_nested_markets=true")
• WebFetch("https://api.elections.kalshi.com/trade-api/v2/markets/candlesticks?tickers=...&start_ts=...&end_ts=...&period_interval=60")
• WebFetch("https://api.elections.kalshi.com/trade-api/v2/markets/MARKET_TICKER/orderbook?depth=10")
• WebFetch("https://api.elections.kalshi.com/trade-api/v2/events?status=open&with_nested_markets=true&limit=50")
• All Kalshi endpoints are public and need no auth headers.

Notify

Send via ./notify (under 4000 chars). Emit only the section(s) for the branch(es) that ran.

• Polymarket only — send the Polymarket report from P4.
• Kalshi only — send the Kalshi report from K6.
• Both — send one combined message: the Kalshi report (K6) first (it's the ranked, decision-oriented view), then a — — — divider, then the Polymarket report (P4). Lead with a one-line cross-venue verdict, e.g. prediction markets — ${today}: [dominant theme across both, or "all quiet both venues"].

If the combined report exceeds the budget, trim in this order: (1) drop Kalshi's "Trending (not tracked)" block, (2) truncate Kalshi events from the bottom of the ranked list, (3) drop Polymarket comment lines, (4) truncate Polymarket events from the bottom.

Notify only on signal. If neither branch found anything worth flagging (no >5pp Polymarket moves, no Kalshi alerts, no themes), a one-line "all quiet" is acceptable signal — but a fully empty/no-change run should send nothing. An explicit all quiet — N events tracked, no moves is useful; an empty template is not.

Log

Append to memory/logs/${today}.md under a single ## Monitor Prediction Markets heading, with a bullet group for each platform that ran:

## Monitor Prediction Markets
- **Platform(s):** both | polymarket | kalshi   (selector: `${var}`)

### Polymarket        (only if the Polymarket branch ran)
- **Events monitored:** N
- **Markets tracked:** N (M open, K closed)
- **Biggest mover:** "[question]" — X% → Y% (+/-Zpp)
- **Alert markets (>5pp move):** [list or "none"]
- **Top comment:** "[excerpt]"

### Kalshi            (only if the Kalshi branch ran)
- **Events monitored:** N (watchlist=W, discovered=D)
- **Markets tracked:** N (M open, K skipped)
- **Top mover:** "[title]" — X% → Y% (Δpp, move_score=S, vol=$V, spread=Sp)
- **Alerts (>5pp, non-thin):** [count; list titles or "none"]
- **Continued-from-yesterday (demoted):** [count]
- **Trending untracked:** [1–2 tickers or "none"]
- **Sources:** events=[status] candlesticks=[status] orderbook=[status]
- **Status:** MONITOR_KALSHI_OK | MONITOR_KALSHI_DEGRADED | MONITOR_KALSHI_NO_CONFIG | MONITOR_KALSHI_ERROR

- **Notification sent:** yes | no

If a market moved dramatically — Polymarket >5pp, or Kalshi >10pp on a non-thin book — or a new category/trend is heating up across multiple events, add a one-line note in memory/MEMORY.md (under a "Prediction market signals" section) for future reference.

Network note

Both branches only fetch public APIs (mode: read-only), so there are no secret-bearing calls. curl works — there is no network sandbox; use WebFetch as a fallback for a flaky public GET (per-platform endpoint lists are in the Polymarket Network note P5 and the Kalshi Network note K9). Never write to the repo beyond memory/logs/ (and an optional memory/MEMORY.md note); produce all output via ./notify and memory/.