claude-for-crypto-research - GitLite

KO | EN

gitlite — search

#hacktoberfest #react #go #golang #android #machine-learning #docker #macos #cli #pytorch #mcp #ios #openai #open-source

claude-for-crypto-research

(★ 105)

This is a simple implementation for crypto research agent inspired by Claude Skills repo

File Explorer

Download Latest Version (.zip)

- - competitive-analysis.md
  - fact-check.md
  - market-trends.md
  - research.md
  - summarize.md
- - data_analyst.txt
  - lead_agent.txt
  - report_writer.txt
  - researcher.txt
- - message_handler.py
  - subagent_tracker.py
  - transcript.py
- agent.py
- polymarket_research_20260128.md
.env.example
.gitignore
pyproject.toml
README.md

# Installation Guide

1. Get the code

git clone https://github.com/juntoku9/claude-for-crypto-research

Downloads the entire project code from GitHub to your computer.

cd claude-for-crypto-research

Moves into the project folder you just downloaded.

2. Python

Easy Recommended

Prerequisites

Git Needed to download the project code from GitHub.
Python 3 On Windows, be sure to check 'Add Python to PATH' during installation.

pip install .

Installs the package published on PyPI directly — no need to clone the source.

python <실행할 파일명>.py # README에서 정확한 실행 파일명을 확인하세요

Runs the Python script (or module).

✅ If it runs without errors and prints output in the terminal, it worked.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Crypto Research Agent

A multi-agent crypto research system that coordinates specialized subagents to research crypto projects, tokens, DeFi protocols, and market trends. Generates comprehensive Markdown reports with airdrop participation guides, market analysis, and Twitter sentiment.

## Quick Start

```bash
# Install dependencies
uv sync

# Set your API keys
export ANTHROPIC_API_KEY="your-anthropic-api-key"
export TWITTER_API_KEY="your-twitter-api-key"

# Or copy .env.example to .env and fill in your keys
cp .env.example .env

# Run the agent (choose mode)
uv run python research_agent/agent.py --mode fast   # Fast mode (2-3 min)
uv run python research_agent/agent.py --mode deep   # Deep mode (8-12 min, default)
uv run python research_agent/agent.py               # Default is deep mode
```

Then ask: "Research Arbitrum" or "Research OpenMind token"

### API Keys Required

- **Anthropic API Key**: Get from [console.anthropic.com](https://console.anthropic.com/settings/keys)
- **Twitter API Key**: Get from [twitterapi.io](https://twitterapi.io) for Twitter/X sentiment analysis

## Research Modes

### Fast Mode (2-3 minutes)
```bash
uv run python research_agent/agent.py --mode fast
```
- **Speed**: 2-3 minutes per research
- **Researchers**: 1-2 agents spawned
- **Searches**: 3-4 essential searches per researcher
- **Twitter**: 10 high-quality tweets (50+ likes, verified)
- **Coverage**: Essentials (overview, price, sentiment, airdrops)
- **Use case**: Quick overview, price check, airdrop opportunities

### Deep Mode (8-12 minutes) - DEFAULT
```bash
uv run python research_agent/agent.py --mode deep
uv run python research_agent/agent.py  # Same as deep
```
- **Speed**: 8-12 minutes per research
- **Researchers**: 3-4 agents spawned in parallel
- **Searches**: 6-8 comprehensive searches per researcher
- **Twitter**: 15-20 tweets with broader coverage
- **Coverage**: Comprehensive (tech, tokenomics, TVL, competitive analysis)
- **Use case**: Investment decisions, thorough due diligence

## How It Works

1. **Lead Agent** breaks your request into 2-4 subtopics
2. Spawns **Researcher** subagents in parallel to search the web and Twitter
3. Each Researcher saves findings to `files/research_notes/`
4. Spawns **Report Writer** to create final Markdown report in `files/reports/`

## Agents

| Agent | Tools | Purpose |
|-------|-------|---------|
| **Lead Agent** | `Task` | Coordinates research, delegates to subagents |
| **Researcher** | `WebSearch`, `Write`, `Skill`, `Bash` | Gathers info from web and Twitter, uses site-specific searches |
| **Report Writer** | `Write`, `Glob`, `Read` | Creates Markdown reports with participation guides |

## Slash Commands

| Command | Description |
|---------|-------------|
| `/research <topic>` | Start focused research on any topic |
| `/competitive-analysis <company>` | Analyze companies or products |
| `/market-trends <industry>` | Research industry trends |
| `/fact-check <claim>` | Verify claims and statements |
| `/summarize` | Summarize all current research findings |

## Example Queries

**Fast Mode** (quick checks):
- "Research OpenMind token"
- "What's the sentiment on Arbitrum?"
- "Check if there's an airdrop for Gradient"

**Deep Mode** (thorough research):
- "Research Arbitrum L2 ecosystem"
- "Compare Aave vs Compound DeFi protocols"
- "Research the AI x Crypto narrative"

**Slash Commands** (any mode):
- `/research Solana`
- `/competitive-analysis Uniswap`
- `/market-trends DeFi`

## Output Structure

```
files/
├── research_notes/     # Markdown files from researchers
└── reports/            # Final Markdown reports with participation guides

logs/
└── session_YYYYMMDD_HHMMSS/
    ├── transcript.txt      # Human-readable conversation
    └── tool_calls.jsonl    # Structured tool usage log
```

## Subagent Tracking with Hooks

The system tracks all tool calls using SDK hooks.

### What Gets Tracked

- **Who**: Which agent (RESEARCHER-1, DATA-ANALYST-1, etc.)
- **What**: Tool name (WebSearch, Write, Bash, etc.)
- **When**: Timestamp
- **Input/Output**: Parameters and results

### How It Works

Hooks intercept every tool call before and after execution:

```python
hooks = Hooks(
    pre_tool_use=[tracker.pre_tool_use_hook],
    post_tool_use=[tracker.post_tool_use_hook]
)
```

The `parent_tool_use_id` links tool calls to their subagent:
- Lead Agent spawns a Researcher via `Task` tool → gets ID "task_123"
- All tool calls from that Researcher include `parent_tool_use_id = "task_123"`
- Hooks use this ID to identify which subagent made the call

### Log Output

**transcript.txt** - Human-readable:
```
[RESEARCHER-1] → WebSearch
    Input: query='quantum computing 2025'
[DATA-ANALYST-1] → Bash
    Input: python matplotlib chart generation
```

**tool_calls.jsonl** - Structured JSON:
```json
{"event":"tool_call_start","agent_id":"RESEARCHER-1","tool_name":"WebSearch",...}
{"event":"tool_call_complete","success":true,"output_size":15234}
```