odds-chart

name: odds-chart description: Plot probability-over-time chart for a Polymarket prediction market using cached history data.

Prediction Market Chart

Use this skill when the user asks to chart, plot, graph, or visualize prediction market odds or probability over time.

Only Polymarket supports price history (CLOB timeseries). Kalshi does not expose historical candlestick data publicly.

Step 1 — Get the history data

If you don't already have market history, call:

prediction_market(query="Get market history for <event name or slug>")

The synthesis output will end with a line like:

(Full raw data: /path/to/workspace/cache/prediction_market_20260227_143012_123456.json)

Copy that path exactly — you'll need it in Step 2.

Step 2 — Read the cache file

read_file("/path/to/cache/prediction_market_....json")

Navigate to the points array based on which command was used:

Each point has the shape: {"timestamp": "2026-02-01T12:00:00Z", "probability": 0.6523}

Also extract:

• Title: result["detail"]["title"] (for market_history/top_mover) or the user's query text (for standalone history)
• Interval: result["history"]["interval"] or result["interval"] (e.g. "1w")

Step 3 — Ensure the charts directory exists

exec("mkdir -p {workspace}/charts")

Replace {workspace} with the actual workspace path from the system prompt.

Step 4 — Write the plotting script

Write the script below to {workspace}/charts/plot_pm.py, substituting the three placeholders:

import json
import sys
from datetime import datetime, timezone

import matplotlib
matplotlib.use("Agg")
import matplotlib.pyplot as plt
import matplotlib.dates as mdates

CACHE_FILE = "{CACHE_FILE_PATH}"
OUTPUT_FILE = "{OUTPUT_PATH}"
TITLE = "{MARKET_TITLE}"

with open(CACHE_FILE) as f:
    raw = json.load(f)

# Locate the points array (handle market_history / top_mover / standalone history)
hist = raw.get("history") or raw
points = hist.get("points", [])
if not points:
    print("No data points found in cache file"); sys.exit(1)


def _parse_ts(val) -> datetime:
    """Handle ISO strings ('2026-02-01T12:00:00Z') and numeric Unix timestamps
    (int, float, or numeric strings in seconds or milliseconds)."""
    if isinstance(val, (int, float)):
        unix = val / 1000 if val > 1e10 else val
        return datetime.fromtimestamp(unix, tz=timezone.utc)
    s = str(val).strip()
    try:
        return datetime.fromisoformat(s.replace("Z", "+00:00"))
    except ValueError:
        pass
    try:
        unix = float(s)
        unix = unix / 1000 if unix > 1e10 else unix
        return datetime.fromtimestamp(unix, tz=timezone.utc)
    except (ValueError, OSError) as exc:
        print(f"Cannot parse timestamp {val!r}: {exc}"); sys.exit(1)


timestamps = [_parse_ts(p["timestamp"]) for p in points]
probs = [round(float(p["probability"]) * 100, 2) for p in points]

# ── Chart ─────────────────────────────────────────────────────────────
fig, ax = plt.subplots(figsize=(12, 5))
fig.patch.set_facecolor("#0f1117")
ax.set_facecolor("#0f1117")

ax.plot(timestamps, probs, color="#00d4aa", linewidth=1.8, zorder=3)
ax.fill_between(timestamps, probs, alpha=0.12, color="#00d4aa")

# Annotate start and end probability
ax.annotate(f"{probs[0]:.1f}%", xy=(timestamps[0], probs[0]),
            xytext=(6, 4), textcoords="offset points",
            color="#00d4aa", fontsize=9)
ax.annotate(f"{probs[-1]:.1f}%", xy=(timestamps[-1], probs[-1]),
            xytext=(-30, 4), textcoords="offset points",
            color="#00d4aa", fontsize=9)

ax.set_ylim(max(0, min(probs) - 5), min(100, max(probs) + 5))
ax.yaxis.set_major_formatter(plt.FuncFormatter(lambda y, _: f"{y:.0f}%"))
ax.xaxis.set_major_formatter(mdates.DateFormatter("%b %d"))
fig.autofmt_xdate(rotation=30, ha="right")

ax.set_title(TITLE, color="white", fontsize=13, pad=12, wrap=True)
ax.set_ylabel("Yes Probability", color="#aaaaaa", fontsize=10)
ax.tick_params(colors="#888888")
for spine in ax.spines.values():
    spine.set_edgecolor("#2a2a2a")
ax.grid(axis="y", color="#1e1e1e", linewidth=0.8)

plt.tight_layout()
plt.savefig(OUTPUT_FILE, dpi=150, bbox_inches="tight",
            facecolor=fig.get_facecolor())
plt.close()
print(f"Chart saved: {OUTPUT_FILE}")

Substitutions:

• {CACHE_FILE_PATH} → the exact cache file path from Step 2
• {OUTPUT_PATH} → {workspace}/charts/pm_{slug}_{YYYYMMDD}.png (use the event slug or a short title slug and today's date)
• {MARKET_TITLE} → the event title string

Step 5 — Execute the script

exec("python3 {workspace}/charts/plot_pm.py")

If it exits with "Chart saved: ...", proceed to Step 6. If it errors, check that the cache file path is correct and the points array is non-empty.

Step 6 — Report the result

Tell the user:

1. The chart was saved to {output_path} (give the full path so they can open it)
2. A one-sentence trend summary, e.g.:

   "Probability climbed from 34% to 61% over the past week, with the sharpest move on Feb 24."

Use the summary object from the cache file (start.probability, end.probability, change_pct) to fill in those numbers — no need to read individual points.

Notes

• Downsampled data: if downsampled: true in the cache, the 30 evenly-spaced points are sufficient for a clean chart. Full resolution isn't needed.
• Multi-sub-market events: market_history always charts the first sub-market's Yes token. If the user wants a different sub-market, they need the specific token_id (from market_detail → clob_token_ids[0]) and the standalone history command.
• Kalshi: has no public price history endpoint — tell the user only Polymarket charts are supported.