Bitget Agent Hub is Bitget's official open-source AI Agent ecosystem. This repository is bitget-agent-skill — the skill surface that teaches terminal AI agents when and how to correctly invoke Bitget tools. Ecosystem packages: agent-skill (this repo), agent-cli, agent-mcp, agent-sdk, bitget-signal, agent-hub.

@bitget-ai/bitget-agent-skill is Bitget's official AI trading skill file, giving Claude Code, Codex, and OpenClaw the judgment to use Bitget tools correctly — knowing when to invoke Bitget, how to break your intent into the right multi-step operations, how to handle confirmation before write actions, and what to do when something goes wrong. Pure markdown, runtime-free.

Is this for you? You're already using Claude Code, Codex, or OpenClaw and want it to actually understand your trading instructions — not just have the tool but lack the reasoning to use it — install this.

Quick Start

Paste the block below into your AI agent — it will handle the entire setup (requires Node.js 20+):

Please help me install the Bitget trading skill (requires Node.js 20+): 1. Run npx @bitget-ai/bitget-agent-skill --target all to deploy the skill files to my AI tool; 2. Run npm install -g @bitget-ai/bitget-agent-cli to install the bgc CLI globally. Tell me the result when done.

Both must be installed together: the skill file is the AI's instruction manual, bgc is the execution tool — the skill teaches the AI when to call it and how to build commands, bgc actually makes the API requests. Without either one, the full flow is incomplete.

Verify it worked. Ask your AI: "What trading modules does Bitget support?" If it returns the correct module list (market, trade, account, etc.), the skill is loaded and active.

What Changes After Installing

graph LR
    A[You<br/>Natural Language] --> B[Claude Code / Codex<br/>with bitget-agent-skill]
    B -->|Generates v3 Command| C[bgc CLI<br/>bitget-agent-cli]
    C -->|HMAC-SHA256 Signed| D[Bitget REST API]
    style B fill:#28a745,stroke:#333,stroke-width:2px,color:#fff

Without the skill, bgc executes one command at a time. With the skill loaded, the AI can chain your intent into a sequence of operations across modules — check positions, evaluate conditions, place the order, handle the result — all in one go.

• Trigger recognition in English and Chinese: whether you say "buy BTC at market" or「帮我买 BTC」, the AI recognizes it as a Bitget trading instruction and calls the right tool automatically — no need to specify command syntax
• v3 grammar and full reference: the skill file includes the bgc v3 grammar (bgc <tool> --action <name>) and an auto-generated catalog of all 109 operations — the AI knows which intent verb maps to which operation and how to fill in the parameters
• Multi-step workflow: "check my positions, and if I have no BTC open a 10x long" — the AI queries first, reasons, then places the order, showing [CAUTION] before every write operation and waiting for your confirmation
• Paper trading and error handling: tell the AI "use demo account" to switch to Bitget's sandbox automatically; when API errors occur, the AI knows whether to retry or surface the issue to you — instead of getting stuck

Why a Skill (vs. just installing bgc)

The CLI alone gives the AI 16 intent verbs over the Unified Trading Account (109 operations) but no semantics — the assistant doesn't know when to reach for Bitget, which verb to use, or how to drive it safely.

The skill adds:

• A trigger description so Claude Code / Codex / OpenClaw automatically invokes the skill on any Bitget-relevant request — including casual phrasings ("buy 0.1 BTC at market") and Chinese ("查看我的持仓").
• The v3 grammar (bgc <tool> --action <name> --<param> <value>) and a discover-first workflow, plus a complete catalog of every verb, action, and parameter auto-generated from the SDK.
• Operational guidance for write-safety (--dry-run / --confirm), close-direction rules, demo trading, error categories, and auth setup.

Result: you say "buy 0.1 BTC at market on Bitget" and it actually works — the assistant composes bgc order --action place --category SPOT --symbol BTCUSDT --side buy --orderType market --qty 0.1, previews it with --dry-run, confirms with you, then runs it.

What Ships in the Skill

| File | Purpose | |---|---| | SKILL.md | Top-level skill definition with trigger phrases (English + Chinese) so the AI knows when to invoke, plus the v3 grammar, the 16-verb map, and the write-safety workflow. | | references/commands.md | Auto-generated static catalog — every domain, verb, action, and parameter (with enums + descriptions), from the SDK's own discover surface. | | references/discover-guide.md | How to navigate the live surface with bgc discover (the four rungs, search, raw, hidden modules). | | references/trading-safety.md | Close-direction rules (one-way/hedge), TP/SL, qty units, cancel-all, and withdrawal safety. | | references/auth-setup.md | How to create a Bitget API key and configure environment variables. | | references/demo-trading.md | How to use --paper-trading for safe rehearsal. | | references/error-codes.md | The v3 error payload, category-driven recovery, and the curated Bitget code table. |

Installation

Step 1: Install the CLI

npm install -g @bitget-ai/bitget-agent-cli

Why -g? bgc is a persistent CLI your AI calls dozens of times per session — it must be on $PATH. This skill, by contrast, is a one-shot deploy: npx copies markdown into your AI tool's skill directory and exits. npx always pulls the latest version, so subsequent runs auto-upgrade you.

Step 2: Deploy the Skill File

# Default — Claude Code only
npx @bitget-ai/bitget-agent-skill

# All supported AI tools
npx @bitget-ai/bitget-agent-skill --target all

# Choose interactively
npx @bitget-ai/bitget-agent-skill --interactive

--target all deploys to Claude Code, Codex, and OpenClaw simultaneously.

| Target | Skill location | |---|---| | Claude Code | ~/.claude/skills/bitget-agent-skill/ | | Codex | ~/.codex/skills/bitget-agent-skill/ | | OpenClaw | ~/.openclaw/skills/bitget-agent-skill/ |

Step 3: Set Up Credentials

export BITGET_API_KEY="your-api-key"
export BITGET_SECRET_KEY="your-secret-key"
export BITGET_PASSPHRASE="your-passphrase"

Create an API key if you don't have one. For demo trading, use a Demo API Key instead:

export BITGET_DEMO_API_KEY="your-demo-key"
export BITGET_DEMO_SECRET_KEY="your-demo-secret"
export BITGET_DEMO_PASSPHRASE="your-demo-passphrase"

Step 4: Restart and Verify

Close and reopen your AI tool, then ask: "What Bitget trading capabilities do you have?"

Related Tools

| Tool | Relationship | |---|---| | @bitget-ai/bitget-agent-cli | Must be installed together: bgc is the execution tool, the skill file is the AI's instruction manual — neither works without the other | | @bitget-ai/bitget-agent-mcp | Alternative for desktop AI (Claude Desktop, Cursor) — no bgc required | | @bitget-ai/bitget-signal | Market analysis skills — no API key needed. This skill handles trading execution; bitget-signal handles market analysis. Complementary, can be installed side by side |

Security

The skill files themselves:

• No API key required — pure Markdown
• No network access — no runtime process
• No credential storage — keys are passed via environment variables to bgc

All actual API calls go through bgc, which provides:

• Credentials live only in environment variables (no .env file parsing)
• --read-only strips all write tools
• --paper-trading switches to the demo environment
• Local HMAC-SHA256 request signing

Troubleshooting

"Skill not recognized by AI"

Cause: Skill file not deployed or AI tool not restarted.

Fix:

1. Re-deploy: npx @bitget-ai/bitget-agent-skill --target all
2. Restart your AI tool completely
3. Verify: ask "What Bitget trading modules are available?"

"AI has bgc but doesn't understand trading commands"

Cause: The skill file is missing — bgc is installed but the AI has no instruction manual.

Fix: Deploy the skill and verify the files landed in the correct location:

| AI Tool | Expected Location | |---|---| | Claude Code | ~/.claude/skills/bitget-agent-skill/SKILL.md | | Codex | ~/.codex/skills/bitget-agent-skill/SKILL.md | | OpenClaw | ~/.openclaw/skills/bitget-agent-skill/SKILL.md |

"bgc: command not found"

Cause: bitget-agent-cli not installed or not on $PATH.

Fix: npm install -g @bitget-ai/bitget-agent-cli then restart your terminal.

Updates

# Update all Bitget AI packages
npx @bitget-ai/bitget-agent-installer upgrade-all --target all

# Or update just this skill
npx @bitget-ai/bitget-agent-skill@latest

FAQ

Will the AI automatically place orders after the skill is installed? No. For write operations like placing orders or transferring funds, the AI displays a [CAUTION] annotation and waits for your confirmation before executing.

How do I trade without real money? Tell your AI "use demo account" or "use paper trading" — it will switch to Bitget's Demo Trading environment automatically. Requires a Demo API Key.

What's the difference between this and the MCP server? Terminal AI (Claude Code, Codex) → install bitget-agent-skill + bgc. Desktop AI (Claude Desktop, Cursor) → install bitget-agent-mcp. Both enable natural language Bitget trading — they adapt to different AI environments.

Does the skill file require an API key or network access? No. Pure Markdown files with no runtime process. All API calls happen at execution time through bgc.

Is it free? MIT-licensed and free to use.

Contributing

• Bug reports & feature requests: GitHub Issues
• Discussions: GitHub Discussions
• Security concerns: [email protected]

License

MIT — Free for personal and commercial use.

Part of the Bitget Agent Hub ecosystem — the official open-source AI Agent ecosystem from Bitget. Trading Stack · Skill Surface. ⚠️ Trading cryptocurrency carries risk; you are responsible for any orders your AI agent places. Use demo account to practice safely before trading with real funds. Foundation: agent-sdk · Execution: agent-cli · Desktop: agent-mcp · Market signals: bitget-signal