Skip to content
BranchesTags

Repository files navigation

TradingAgents Analysis Skill

English | 中文

License Rust

A multi-agent stock and crypto trading analysis skill for AI agents, inspired by TradingAgents and TradingAgents-CN.

RESEARCH AND EDUCATION ONLY. This skill does not constitute financial advice.

Overview

This skill teaches an AI agent to replicate the TradingAgents multi-agent analysis pipeline. It orchestrates a team of specialized sub-agents to produce a structured, well-reasoned investment decision for any given ticker.

Output: one of five ratings (Buy / Overweight / Hold / Underweight / Sell) with detailed reasoning, key risks, and potential catalysts.

Supported markets:

  • US stocks (AAPL, MSFT, TSLA...)
  • China A-shares (600519, 000858...)
  • HK stocks (0700.HK, 9988.HK...)
  • Crypto (BTC-USD, ETH-USD...)

The pipeline runs in six stages:

  1. Analyst Team (parallel sub-agents): Market, Sentiment, News, Fundamentals
  2. Research Debate: Bull vs Bear researchers (structured argumentation, 1-3 rounds)
  3. Research Manager: Judges the debate, produces an investment plan with a preliminary rating
  4. Trader: Converts the plan into a concrete transaction proposal
  5. Risk Debate: Aggressive vs Conservative vs Neutral analysts (1-3 rounds)
  6. Portfolio Manager: Final decision with confidence level and full reasoning

Installation

Via bunx (recommended)

Install the skill with the built-in Rust installer. bunx (or npx) fetches the package and runs a thin launcher that resolves your platform binary and installs the skill into the generic agent directory ~/.agents/skills by default:

bunx trad-skill@latest              # install to ~/.agents/skills (default)

Options:

bunx trad-skill@latest --agent claude      # install to ~/.claude/skills (Claude Code)
bunx trad-skill@latest --agent opencode    # install to ~/.config/opencode/skills
bunx trad-skill@latest --dir <path>        # install into a custom skills dir
bunx trad-skill@latest --dry-run           # print the install plan without writing

npx trad-skill@latest behaves identically if you don't have bun installed. On older npx versions use npx -p trad-skill@latest trad-skill.

Data tool without installing

The same binary also exposes stock / news / fundamentals / sentiment subcommands. Run them directly via bunx (or npx) without installing the skill — handy for one-off data fetches:

bunx trad-skill@latest stock --symbol AAPL
bunx trad-skill@latest news --symbol AAPL
bunx trad-skill@latest fundamentals --symbol AAPL
bunx trad-skill@latest sentiment --symbol AAPL

For AI Agents

If you are an AI agent and need to install this skill, run:

bunx trad-skill@latest

Then restart or open a new session to load the skill. The skill will be available at:

  • ~/.agents/skills/tradingagents-analysis (default; OpenCode, Cline, Cursor, Windsurf, Codex, etc.)
  • ~/.claude/skills/tradingagents-analysis (Claude Code)
  • ~/.config/opencode/skills/tradingagents-analysis (OpenCode global)

Deprecated: npx skills add (vercel-labs/skills CLI)

Deprecated. The third-party npx skills add halfoffive/trad-skill ... flow (via vercel-labs/skills) is superseded by bunx trad-skill@latest. It still works for now but is no longer recommended and may be removed in a future release.

Manual installation (using raw GitHub links)

Copy the skill files directly from the repository raw URLs.

Claude Code (user-level):

mkdir -p ~/.claude/skills/tradingagents-analysis
curl -sL https://raw.githubusercontent.com/halfoffive/trad-skill/main/skills/tradingagents-analysis/SKILL.md -o ~/.claude/skills/tradingagents-analysis/SKILL.md
mkdir -p ~/.claude/skills/tradingagents-analysis/references/prompts
curl -sL https://raw.githubusercontent.com/halfoffive/trad-skill/main/skills/tradingagents-analysis/references/data-sources.md -o ~/.claude/skills/tradingagents-analysis/references/data-sources.md
curl -sL https://raw.githubusercontent.com/halfoffive/trad-skill/main/skills/tradingagents-analysis/references/indicators.md -o ~/.claude/skills/tradingagents-analysis/references/indicators.md
curl -sL https://raw.githubusercontent.com/halfoffive/trad-skill/main/skills/tradingagents-analysis/references/prompts/README.md -o ~/.claude/skills/tradingagents-analysis/references/prompts/README.md
for f in market_analyst sentiment_analyst news_analyst fundamentals_analyst bull_researcher bear_researcher research_manager trader aggressive_risk conservative_risk neutral_risk portfolio_manager china_market_analyst cn_news_analyst; do
  curl -sL "https://raw.githubusercontent.com/halfoffive/trad-skill/main/skills/tradingagents-analysis/references/prompts/${f}.md" -o ~/.claude/skills/tradingagents-analysis/references/prompts/${f}.md
done

Generic / OpenCode (user-level): replace ~/.claude/skills with ~/.agents/skills in the commands above.

Or simply clone/copy the directory:

git clone --depth 1 https://github.com/halfoffive/trad-skill /tmp/trad-skill
cp -r /tmp/trad-skill/skills/tradingagents-analysis ~/.claude/skills/
rm -rf /tmp/trad-skill

Usage

Trigger the skill by asking your AI agent to analyze a ticker:

  • "Analyze AAPL for me"
  • "What do you think about NVDA?"
  • "Run a multi-agent analysis on BTC-USD"
  • "Give me a trading analysis of 0700.HK"
  • "Analyze 600519 with 3 debate rounds"

The agent will orchestrate the full pipeline and produce a structured investment report ending with a summary table:

Field Value
Ticker AAPL
Date 2025-01-15
Rating Overweight
Confidence medium
Market US

China A-share notes

  • Tickers use the 6-digit pure-number format (e.g., 600519, 000858)
  • trad-skill auto-detects the exchange from the 6-digit code prefix (leading 6 → Shanghai .SS; leading 0/3 → Shenzhen .SZ), so you only need to provide the 6-digit number
  • Data source: Eastmoney price APIs by default; use trad-skill stock --source yahoo to switch to the Yahoo channel (A-share codes are auto-mapped to .SS/.SZ)
  • Chinese news and sentiment analysis are supported
  • China-market-specific analyst prompts are used (china_market_analyst.md, cn_news_analyst.md)

HK stock notes

  • Use a 4-5 digit number + .HK suffix (e.g., 0700.HK or 00700.HK; trad-skill accepts both and zero-pads the code to 5 digits internally)
  • Data source: HK prices default to the Eastmoney APIs (secid 116.<5-digit code>); --source yahoo switches to the Yahoo channel
  • Stock Connect targets and HK main-board stocks are supported

Configuration

Parameter Range Default Description
max_debate_rounds 1-3 1 Bull/bear exchanges in the Research Debate
max_risk_discuss_rounds 1-3 1 Risk analyst exchanges in the Risk Debate
output_language English / Chinese match user Language for all reports
market auto-detect — Detected from ticker suffix

Market auto-detection:

  • 6-digit pure number (e.g., 600519, 000858) → A-shares
  • 4/5-digit pure number (e.g., 0700, 09988) → HK stocks (when no .HK suffix)
  • .HK suffix → HK stocks
  • -USD suffix → Crypto
  • Everything else → US stocks

Architecture

+-----------------------------------------------------+
|                   ANALYST TEAM                       |
|  +----------+ +----------+ +--------+ +----------+  |
|  |  Market  | |Sentiment | |  News  | |Fundament.|  |
|  | Analyst  | | Analyst  | |Analyst | | Analyst  |  |
|  +----------+ +----------+ +--------+ +----------+  |
+-----------------------------------------------------+
                         |
                         v
+-----------------------------------------------------+
|              RESEARCH DEBATE (1-3 rounds)            |
|         Bull Researcher  <->  Bear Researcher        |
+-----------------------------------------------------+
                         |
                         v
              +---------------------+
              |  Research Manager   | -> Investment Plan
              +---------------------+
                         |
                         v
              +---------------------+
              |       Trader        | -> Transaction Proposal
              +---------------------+
                         |
                         v
+-----------------------------------------------------+
|              RISK DEBATE (1-3 rounds)                |
|   Aggressive  <->  Conservative  <->  Neutral       |
+-----------------------------------------------------+
                         |
                         v
              +---------------------+
              | Portfolio Manager   | -> FINAL DECISION
              +---------------------+

Project Structure

trad-skill/                        # repo root (meta files + installer)
├── package.json                  # npm entry point (name: trad-skill)
├── bin/trad-skill.js             # thin JS launcher -> Rust binary (install + data)
├── README.md / README_CN.md       # bilingual docs with language switch
├── CHANGELOG.md                   # version history
├── AGENTS.md                      # AI-agent onboarding doc
├── LICENSE                        # Apache 2.0
├── .github/workflows/ci.yml      # CI: fmt + clippy + test + 7-platform build
├── crates/trad-data/              # Rust source (binary name: trad-skill)
└── skills/
    └── tradingagents-analysis/    # the installable skill
        ├── SKILL.md               # Core skill instructions (pipeline, orchestration, output format)
        └── references/
            ├── prompts/           # 14 verbatim agent role prompts
            │   ├── market_analyst.md
            │   ├── sentiment_analyst.md
            │   ├── news_analyst.md
            │   ├── fundamentals_analyst.md
            │   ├── bull_researcher.md
            │   ├── bear_researcher.md
            │   ├── research_manager.md
            │   ├── trader.md
            │   ├── aggressive_risk.md
            │   ├── conservative_risk.md
            │   ├── neutral_risk.md
            │   ├── portfolio_manager.md
            │   ├── china_market_analyst.md
            │   ├── cn_news_analyst.md
            │   └── README.md
            ├── data-sources.md    # Data source catalog (US + CN markets)
            └── indicators.md      # Technical indicator reference

Data Sources

Source Market Provides API Key
Yahoo Finance US/Global Price, fundamentals, news Free
Alpha Vantage US Price, indicators, fundamentals Free tier
FRED US Macro CPI, unemployment, rates Free
Polymarket Global Prediction probabilities Free
StockTwits US Retail sentiment Free
Reddit US Community discussion Free
Tushare A-shares Price, fundamentals Token
AKShare A-shares/HK Price, news, sentiment Free
Baostock A-shares Historical data Free
TDX (通达信) A-shares Technical indicators Free

See references/data-sources.md for the full catalog with fallback strategies.

Data Tool (trad-skill)

Data is fetched by the trad-skill Rust binary, which provides market data (OHLCV + indicators), news, fundamentals, and sentiment in a single compact output suitable for LLM prompt injection. Platform binaries are distributed as the five @trad-skill/<platform> npm packages (installed as optional dependencies), and the thin launcher bin/trad-skill.js resolves the correct one at runtime. The same binary handles both install and data subcommands.

The agent must run trad-skill by its absolute path inside the installed skill directory (e.g. ~/.agents/skills/tradingagents-analysis/bin/<platform>/trad-skill), because a sub-agent's working directory is the user's project, not the skill folder. The skill's SKILL.md instructs the main agent to resolve that path before spawning analysts.

# Stock data: OHLCV tail + pre-computed indicators + optional stats
trad-skill stock --symbol AAPL --start 2023-07-01 --end 2024-06-30 --tail 30 --stats

# Or run directly via bunx (no install needed):
bunx trad-skill@latest stock --symbol AAPL --start 2023-07-01 --end 2024-06-30 --tail 30 --stats

# Or by absolute path (how the agent invokes them):
~/.agents/skills/tradingagents-analysis/bin/<platform>/trad-skill stock --symbol AAPL --start 2023-07-01 --end 2024-06-30 --tail 30 --stats

# China A-share: use the 6-digit code directly (auto-routed to Eastmoney)
trad-skill stock --symbol 600519 --tail 30

# Pick a data channel explicitly — e.g. route US stocks via Eastmoney when Yahoo
# Finance is blocked in your region (returns "未知错误" / 403):
trad-skill stock --symbol AAPL --source eastmoney

# Fetch news (default --limit 8 per source, summaries truncated)
trad-skill news --symbol AAPL --days 7 --limit 8

# Fetch fundamentals (compact key-metrics table + company profile)
trad-skill fundamentals --symbol AAPL

# Fetch sentiment (default --limit 15, --days 7 Reddit window)
trad-skill sentiment --symbol AAPL --limit 15 --days 7

# China A-share & HK fundamentals/news auto-route to Eastmoney (no Yahoo required):
trad-skill fundamentals --symbol 600519
trad-skill news --symbol 600519
trad-skill stock --symbol 0700.HK        # HK OHLCV via Eastmoney
trad-skill news --symbol 0700.HK         # HK news via Eastmoney
# HK sentiment is not supported (StockTwits/Reddit are US-only; Eastmoney
# 千股千评 does not cover HK) — use the US-listed ticker if available.
Subcommand Defaults (compact) Expand flags
stock --tail 30 + --indicators on --stats, --raw, --source yahoo|eastmoney
news --limit 8, 200-char summaries --limit N, --days N
fundamentals compact key-metrics table —
sentiment --limit 15, --days 7 (Reddit), 8 messages/posts shown --limit N, --days N

stock --source selects the data channel: by default US/Crypto use Yahoo Finance and A-shares/HK use Eastmoney, auto-detected from the symbol. Pass --source eastmoney to route US stocks through Eastmoney (handy when Yahoo is region-blocked), or --source yahoo to force Yahoo (A-share/HK symbols are mapped to .SS/.SZ/.HK). Eastmoney does not serve crypto.

When Yahoo Finance is unreachable (symptom: 未知错误, 401 Unauthorized, or 403 Forbidden from datacenter/cloud IPs): A-share stock / fundamentals / news already use Eastmoney automatically — just pass the 6-digit code. For US price data, add --source eastmoney. US fundamentals / news have no Eastmoney equivalent, so the agent falls back to web search for those parts.

trad-skill is the primary data source and is tried first. It is not a hard dependency in the sense that, if a subcommand errors or a source is unavailable, the agent falls back to web search / browser tools only for the parts the command could not provide — it never skips the binary outright.

Credits

Disclaimer

This skill is for RESEARCH AND EDUCATION ONLY.

  • Nothing produced by this skill constitutes financial, investment, or trading advice.
  • Past performance does not guarantee future results.
  • LLM-generated analysis is non-deterministic and may contain factual errors, hallucinated data, or flawed reasoning.
  • Always consult a qualified, licensed financial advisor before making any investment decision.
  • The authors and contributors accept no liability for any losses incurred from acting on this output.

License

Apache License 2.0

About

No description, website, or topics provided.

Resources

Readme
Apache-2.0 license

Contributing

Contributing
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages