TradingAgents Multi-Agent A-Share Investment Research and Analysis

🎯 Quick Start

• "Help me analyze Kweichow Moutai"
• "Is CATL worth buying?"
• "Analyze the technical aspect of 600519"
• "How is BYD's recent capital flow?"

🤖 System Architecture: 5 Stages · 15 Agents

TradingAgents Multi-Agent Investment Research

🤖 System Architecture: 5 Stages · 15 Agents

📋 Applicable Scenarios

• In-depth individual stock analysis (technical + fundamental)
• Investment decision reference
• Post-market review analysis
• Risk assessment of held positions
• Capital flow and market sentiment research and judgment

• Real-time intraday monitoring (analysis takes 1-5 minutes)
• Ultra-short-term trading (minute-level decision making)
• Non-A-share markets such as cryptocurrency and US stocks

🔒 Privacy and Security

• Sending scope: This skill only extracts parameters such as stock name/code, analysis date, analysis perspective from the conversation, and sends them to the backend API as

  symbol

  /

  trade_date

  /

  horizons

  fields. Do not send the original conversation text, do not read local files, and do not upload any other private data.
• Token security:

  TRADINGAGENTS_TOKEN

  (format

  ta-sk-*

  ) is the only credential to access the backend. Please use the minimum privilege token. If you suspect leakage, please immediately revoke and regenerate it at app.510168.xyz.
• Sensitive content prompt: Please do not paste personal account information, real positions or other sensitive content into the analysis request. This skill cannot prevent users from actively submitting these contents.
• Self-hosting: If you need full control over the data flow, you can refer to GitHub Documentation to deploy the backend by yourself, and point

  TRADINGAGENTS_API_URL

  to your self-built server.

About credential metadata: The frontmatter of this skill declares

TRADINGAGENTS_TOKEN

as

primaryEnv

in

metadata.openclaw

and lists it in

requires.env

.

🔒 Privacy & Data Transmission

• What is sent: Only the extracted stock symbol, trade date, and analysis parameters (

  symbol

  ,

  trade_date

  ,

  horizons

  ) are transmitted to the backend. The raw conversation text is never forwarded.
• Token:

  TRADINGAGENTS_TOKEN

  (pattern

  ta-sk-*

  ) is the sole credential. Use a minimal-privilege token and rotate it immediately if compromised.
• Sensitive content: Do not paste personal account data, real positions, or other sensitive information into analysis requests.
• Self-hosting: For full data sovereignty, deploy the backend yourself and set

  TRADINGAGENTS_API_URL

  to your server. See the GitHub repo.

Credential metadata: This skill's frontmatter declares

TRADINGAGENTS_TOKEN

as

primaryEnv

under

metadata.openclaw.requires.env

.

⚙️ Quick Configuration

1. Log in to https://app.510168.xyz
2. Go to Settings → API Tokens to create a token
3. Configure environment variables:

export TRADINGAGENTS_TOKEN="ta-sk-your_key_here"

# 1. Deploy the backend, refer to https://github.com/KylinMountain/TradingAgents-AShare
# 2. Point the API address to the self-built service
export TRADINGAGENTS_API_URL="http://your-server:8000"
export TRADINGAGENTS_TOKEN="ta-sk-your_key_here"

🚀 Common Operations

# Script path (relative to skill directory)
bash scripts/analyze.sh <symbol[,symbol2,...]> [trade_date] [horizons]

# Single analysis
bash scripts/analyze.sh 贵州茅台
bash scripts/analyze.sh 600519.SH 2026-03-22
bash scripts/analyze.sh 600519.SH 2026-03-22 medium

# Batch analysis (comma separated, parallel submission, unified waiting)
bash scripts/analyze.sh 贵州茅台,比亚迪,宁德时代
bash scripts/analyze.sh 600519.SH,002594.SZ,300750.SZ 2026-03-22

• POLL_INTERVAL

  — polling interval in seconds (default 15)

• POLL_TIMEOUT

  — maximum waiting seconds (default 600)

$TRADINGAGENTS_TOKEN

1. Submit analysis task

curl -X POST "${TRADINGAGENTS_API_URL:-https://api.510168.xyz}/v1/analyze" \
  -H "Authorization: Bearer $TRADINGAGENTS_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"symbol": "贵州茅台"}'

1. Query task status

curl "${TRADINGAGENTS_API_URL:-https://api.510168.xyz}/v1/jobs/{job_id}" \
  -H "Authorization: Bearer $TRADINGAGENTS_TOKEN"

1. Get complete analysis results (after the task is completed)

curl "${TRADINGAGENTS_API_URL:-https://api.510168.xyz}/v1/jobs/{job_id}/result" \
  -H "Authorization: Bearer $TRADINGAGENTS_TOKEN"

📊 Sample Output

{
  "decision": "BUY",
  "direction": "看多",
  "confidence": 78,
  "target_price": 1850.0,
  "stop_loss_price": 1680.0,
  "risk_items": [
    {"name": "估值偏高", "level": "medium", "description": "当前 PE 处于历史 75 分位"},
    {"name": "外资流出", "level": "low",    "description": "近 5 日北向资金小幅净流出"}
  ],
  "key_metrics": [
    {"name": "PE",   "value": "32.5x",  "status": "neutral"},
    {"name": "ROE",  "value": "31.2%",  "status": "good"},
    {"name": "毛利率", "value": "91.5%", "status": "good"}
  ],
  "final_trade_decision": "综合技术面突破与基本面支撑，建议逢低分批建仓..."
}

🔄 Task Execution Process

1. Identify the target: Only extract the stock name or code (and optional date/perspective) from the conversation, and do not send the original conversation text
2. Notify the user: Feedback that the task is about to be submitted, and it is expected to take 1-5 minutes
3. Execute the script: Use the Bash tool to run

   bash scripts/analyze.sh <symbol> [date] [horizons]

   (set

   run_in_background: true

   ), and the script automatically completes submission, polling and result acquisition
4. Summarize the conclusion: After the script output is completed, parse the JSON results, and show the user the decision, direction, target price, and risk points

Important: Do not manually write curl polling loops, directly use the

scripts/analyze.sh

script.

📌 Supported Target Range

• Shanghai and Shenzhen A-shares: Chinese name (such as "BYD", "CATL") or code (

  002594.SZ

  ,

  601012.SH

  )

💡 Notes

• Polling frequency: The interval between each polling is not less than 15 seconds
• Data robustness: If some data sources are missing, the system will conduct spillover analysis based on macro and industry logic
• Short-term mode: When entering "Analyze XX short-term", the system automatically switches to 14-day technical analysis, skips financial report data, and is faster