Official descriptionOpen source3d ago

superwrldcoin/trading-agent

Analysis-only 4H trading research agent (Claude Code): data tools, skills, memory protocol

View on GitHub ↗Direct link · no redirect page

Star count: 1 ★ · superwrldcoin/trading-agent · read from the GitHub search API · (checked 2026-10-16 20:00:20)
Source: read directly from its GitHub page · type: open-source project
Harvest history: first seen 2026-10-14 · last seen 2026-10-16 · seen 3× · listed by 1 sources
Score: score 103 · 1 ★ · 1 sources · repo

FAQ (answers come from our records)

How do I install it?

We did not find an install command for this repository (our install coverage is only about 4%). Check its README or releases page.

Who is it for?

We filed it under these tags: 开源项目. The tags come from the entry itself, not a taxonomy we imposed.

Who maintains it?

repository superwrldcoin/trading-agent；first catalogued 2026-10-14；last seen 2026-10-16。There is no official certification or quality rating here — we record sources and dates only.

Third-party quality signals

trading-agent

以下摘录自该仓库的 README（片段） · 在 GitHub 读完整 README →

An **analysis-only** trading research agent built on Claude Code. It reads 4H market structure off
previous day/week/month liquidity levels (PDH/PDL, PWH/PWL, PMthH/PMthL), grades conviction with
**EMA + VWAP**, and checks position math (R:R, liquidation, sizing), all from live public data.
It never places trades and never asks for or stores API keys.
Watchlist (from memory/universe.yaml): BTC/USDT, BCH/USDT (OKX, yfinance backup), XAU/USD (OKX Tether Gold proxy,
cross-checked against spot), SI (COMEX silver futures), SLV, GLD, MSFT (yfinance), plus 5 macro drivers and 8 pair instruments.
> Not financial advice. Outputs are analysis; you make and execute every decision.
Ask it something
| Interface | How | What runs |
|---|---|---|
| Web page | double-click start-ui.cmd (or python app/server.py), then open http://127.0.0.1:8765 | **Quick check** (no AI, ~30s on first fetch) or **Full agent** (Claude Code with every skill and memory file, a few minutes). Live step progress, cancel, run history, price ladder, settings drawer (theme, animations, defaults, shortcuts) |
| Terminal | .\ask.ps1 "BTC long 20x entry 98,400" / ./ask.sh ..., add -Agent / --agent for the full agent | same |
| Claude Code | cd trading-agent && claude, then type the question | the agent itself (interactive) |
The **quick check** parses the request, fetches 15M/1H/4H/1D data, and returns:
- a verdict
- EMA + VWAP conviction for the side
- structure
- a stop and targets derived from structure (or yours)
- a milestone table with R, P&L/ROE, and liquidation
- flags, such as liquidation hit before the stop, an entry far from the market, or R:R under 2
The **full agent** runs the quick check, then applies the skills, adds the event check and a
P(T1 before stop) judgment, and logs the session.
The full agent needs the Claude Code CLI logged in on the machine.
Run claude once inside this folder and accept the trust prompt. No API key is stored.
Everyday commands
After a trade closes, in claude: post-mortem: BTC long, entry ..., exit ..., size ..., stop ..., 20x, opened ..., closed ..., thesis was "...".
Review the proposed MEMORY_UPDATE, approve it, then commit memory/journal.md. Trades, sessions, and positions stay private (gitignored).
Setup
Layout
| Path | What |
|---|---|
| CLAUDE.md, AGENT.md | Agent rules: role, write scope, tools and fallbacks, session workflow, report format |
| skills/ | Procedures with worked examples: report-format, levels-and-entries, multi-timeframe-momentum (EMA + VWAP conviction), market-structure, risk-and-sizing, leveraged-position-math, leverage-volatility-check, rules-check, trade-plan, portfolio-exposure, stress-test, options-greeks, bear-case-review, data-ingest, macro-and-catalysts, post-mortem |