Skip to content

Examples & agent skills

Terminal window
git clone https://gitlab.com/cotes-ai-public/spectraledge-examples.git
cd spectraledge-examples
export SPECTRALEDGE_API_KEY=cotes_sk_...
Sample Language Does
python/stream_client.py Python Prints contest.opened, signal.emitted and contest.settled events; reconnects with backoff
node/stream.ts TypeScript / Node The same with fetch and no runtime dependencies
python/discord_alerts.py Python Posts each new signal to a Discord webhook, with an optional --min-confidence
python/paper_trade_logger.py Python Logs each signal next to Kalshi’s current ask to CSV; never places orders
Terminal window
cd python && uv run stream_client.py # Python 3.10+
cd node && npm start # Node 22.18+

The stream clients follow the rules in Streaming (SSE): one stream instead of polling, backoff from 1 s to 30 s with jitter, Last-Event-ID resume, a reconnect after 45 s without bytes, and a clean exit on 401, 402 or the in-stream unauthorized event. spectraledge_stream.py and sse.ts are importable on their own.

agent-skills/ is a Claude Code plugin (spectraledge-signals) with five skills. The analysis runs in a bundled standard-library Python script, so hundreds of contests never pass through the model’s context.

Skill Answers
spectraledge-contest-readings What did the model say through the last, current or a given contest?
spectraledge-threshold-backtest At my confidence threshold, how often is the first qualifying reading right, and does it beat its entry price?
spectraledge-calibration-check Does 80% mean 80%, and is the model closer to the outcome than the market price?
spectraledge-edge-backtest Do positive-edge readings pay off, bucket by bucket?
spectraledge-live-watch Tell me when a live reading crosses my threshold, and how it settled.

Claude Code:

/plugin marketplace add https://gitlab.com/cotes-ai-public/spectraledge-examples.git
/plugin install spectraledge-signals@spectraledge

claude.ai and the Claude apps: Customize → Plugins → Add → Add marketplace, paste https://gitlab.com/cotes-ai-public/spectraledge-examples.git, then install spectraledge-signals.

Other agents: copy the whole agent-skills/ folder, keeping skills/ and scripts/ side by side.

The plugin registers the MCP server. Sign in there, or set SPECTRALEDGE_API_KEY so the script fetches from the API itself. spectraledge-live-watch needs the API key, since it reads the event stream.

Backtests report P&L before Kalshi fees, and every skill asks for your confidence threshold before counting a hit rate.

cotes-ai-public/companion-indicators (MIT) has free TradingView indicators for KXBTC15M: a window and strike overlay and a distance-to-strike pane. They don’t call the API.

Issues and merge requests are welcome on GitLab. Questions go to Discord.