GetAgent Skill

Author, backtest, and ship algorithmic crypto trading strategies on Bitget — through your AI coding agent.

GetAgent is Bitget's managed platform for AI-authored quantitative trading strategies (Playbooks). This Agent Skill teaches your AI coding agent — Claude Code, Cursor, or Codex — the full end-to-end GetAgent workflow: writing strategy code in Python, validating it locally, uploading to the managed sandbox, running historical backtests, publishing accepted versions, and opening subscriptions.

Quick Start

npx @bitget-ai/getagent-skill install --client claude

Then open Claude Code and ask:

Build me a momentum strategy on BTC perp using EMA crossover and run a 30-day backtest.

The agent scaffolds a Playbook package, writes the strategy code, validates it locally, uploads it to GetAgent Cloud, runs the backtest in the managed sandbox, and shows you results — all without leaving the chat.

For other agents:

npx @bitget-ai/getagent-skill install --client cursor   # installs to ~/.cursor/skills/getagent
npx @bitget-ai/getagent-skill install --client codex    # installs to ~/.codex/skills/getagent
npx @bitget-ai/getagent-skill install --client all      # all three

How It Works

GetAgent splits authoring from execution:

• Local (your machine, via this skill) — write strategy Python, validate package structure, call the Playbook control-plane API through your agent.
• Managed sandbox (GetAgent Cloud) — the real getagent.data, getagent.trade, getagent.llm, and getagent.backtest modules execute with live market data and brokered exchange access.

You write code against the documented getagent.* API. When your agent uploads the Playbook, GetAgent provisions a matching sandbox image and runs the code there, not on your laptop. API keys, market data licensing, and execution risk all stay inside the managed boundary.

A typical Playbook entry point looks like:

from getagent import backtest, data, runtime

def run() -> None:
    bars = data.crypto.futures.kline(symbol="BTCUSDT", interval="1h", limit=1000)
    frame = backtest.prepare_frame(bars)
    # ... your strategy emits signals via runtime ...
    runtime.emit_signal(action="buy", symbol="BTCUSDT", confidence=0.8)

A complete reference strategy lives in skills/getagent/examples/btc-ema-cross-demo/.

What's In The Skill

• SKILL.md — agent instructions, conversation templates, control flow
• references/ — package schema, API reference, sandbox runtime, and authoring docs
• scripts/validate.py — static package validator (runs locally before upload)
• examples/ — runnable example Playbooks

Requirements

• Node.js 18+ to install the skill via npm
• Python 3.11+ for local static validation
• A GetAgent Cloud account and API credentials — your agent will guide authentication on first use

Versioning

This package follows Semantic Versioning:

• Patch (v0.1.x) — bug fixes, doc updates, reference regen with no API change
• Minor (v0.x.0) — additive: new endpoints, new examples, new authoring surfaces
• Major (vX.0.0) — breaking changes to the documented authoring API

If the skill documents an API, the sandbox image GetAgent Cloud provisions for your Playbook contains that API.

License

Proprietary. See package.json.

Support

• Issues: https://github.com/Bitget-AI/getagent-skill/issues
• Skill content: open ~/.claude/skills/getagent/SKILL.md (or the matching path for cursor/codex) after install