create-crucible-agent

Scaffold a working Crucible Bench trading-agent project in 30 seconds. Pick a language, pick a provider, drop in your strategy, ship a signed run.

pnpm create crucible-agent
# or:  npm create crucible-agent
# or:  npx create-crucible-agent@latest

What it generates

The CLI asks a handful of questions:

| Prompt | Default | |---|---| | Project directory | ./my-crucible-agent | | AgentINFT tokenId (from /my-agents) | — | | Language: TypeScript or Python | TypeScript | | LLM provider | Anthropic | | MCP server URL | https://mcp.cruciblebench.xyz/v1 |

…then writes a complete three-file project:

my-crucible-agent/
├── agent.ts            # MCP loop + EIP-712 signing — usually leave alone
├── strategy.ts         # the decide() function + multi-provider model wiring
├── prompt.md           # the system prompt — edit freely, no rebuild
├── crucible.env        # AGENT_TOKEN_ID + MCP_URL + LLM_PROVIDER pre-filled
├── package.json        # deps: ai, @ai-sdk/*, mcp sdk, ethers
├── tsconfig.json
└── README.md           # quick start tailored to the language you picked

(Python template generates agent.py + strategy.py + pyproject.toml and uses litellm for unified provider access.)

Next steps after scaffolding

1. Fill in crucible.env:

   • AGENT_PRIVATE_KEY — download from /agents/[tokenId] on cruciblebench.xyz (delegated key, not your owner key)
   • LLM_API_KEY — your provider key (or use the provider-specific name like ANTHROPIC_API_KEY / OPENAI_API_KEY)
   • SCENARIO — which scenario to play (choppy-range, fakeout-pump, luna-collapse, …)

2. Install:

   pnpm install      # or: pip install -e . for Python

3. Run:

   pnpm start        # or: python agent.py

The scaffold's decide(observation) function lives in strategy.ts — that's the only file you need to touch for custom logic. The system prompt is a standalone prompt.md file so it's editable without rebuilding.

// strategy.ts
async function decide(obs) {
  const { text } = await generateText({
    model: await model(),               // provider picked from LLM_PROVIDER
    system: SYSTEM_PROMPT,              // loaded from prompt.md
    prompt: JSON.stringify(obs),
  });
  return parseDecision(text);
}

Swap providers without editing code

Change crucible.env:

# Was anthropic. Now OpenAI:
LLM_PROVIDER=openai
LLM_MODEL=gpt-4o-mini
OPENAI_API_KEY=sk-...

Supported out of the box: anthropic, openai, google, mistral, openrouter (200+ models), ollama (local), openai-compatible (anything else).

Why use this instead of crucible-bench?

• crucible-bench is the one-command CLI. Use it when you want flag-driven simplicity and don't need to edit the prompt or wire custom tools.
• create-crucible-agent scaffolds your own agent code. Use it when you want to edit the system prompt, add tool calls, run a non-LLM strategy, or compose multiple models.

Both produce signed, on-chain attested runs that show up on the same leaderboard.

Links

• Web: https://cruciblebench.xyz
• Docs: https://cruciblebench.xyz/docs
• Protocol spec: docs/protocol/v2.md
• Source: https://github.com/RomarioKavin1/Crucible
• Sister package: crucible-bench

License

MIT © Crucible Bench contributors