drt-self-improving-agent

这是一款本地运行的DRT自改进交易代理，核心功能包括：记录每笔交易的DRT类型、盈亏等信息，分析交易盈亏模式与规律，生成针对性优化规则，还能提醒优质交易时段。适合交易者每次交易后通过CLI提交数据，定期查看分析结果调整策略，所有操作无需联网，输出直观规则建议助力提升胜率。

northcap-group

热度 36.22

最近更新 2026-09-01 05:03:36

分类：数据依赖：代码执行 / 读写文件应用场景：财务 / 数据处理 / 报告撰写

安装技能

复制下面的话术发给你的 Agent，它会通过 secure-skills 检索并安装该技能。

执行命令安装此 skill： npx secure-skills-cli add drt-self-improving-agent

Metadata

name: drt-self-improving-agent
description: Self-improving DRT/ICT trading agent — journals every trade (setup type, bias, R:R, outcome), analyzes its own win/loss patterns, and builds a personal trading memory that makes the agent smarter over time. 100% lokal — ingen netværkskald.
metadata: {"clawbot":{"requires":{"python3":true},"notes":"100% lokal analyse — ingen netværkskald, ingen API-nøgle."}}

DRT Self-Improving Trading Agent 🤖📈

A trading agent that learns from its own trades. Every trade is journaled → analyzed → patterns are discovered → the agent adjusts its own rules. The more trades, the sharper it gets.

What the skill does

Journals every trade (DRT type, bias, entry/SL/TP, R:R, outcome, killzone)
Analyzes patterns — which setups win, which lose, what time of day
Builds memory — trades.json grows with every trade
Learns and adjusts — prints changed rules based on data (not gut feeling)
Proactive — reminds about killzones and A+ setups

Files

drt-self-improving-agent/
├── SKILL.md
├── scripts/
│   ├── journal.py      # Add trade to trades.json (CLI)
│   ├── analyze.py      # Analyze patterns + print learnings
└── data/
    └── trades.json     # Trade memory (auto-created)

Quick start

# Journal a trade (after every trade!)
python3 scripts/journal.py --symbol SP500 --bias LONG --type 2 \
  --entry 7741 --sl 7681 --tp 7802 --rr 2.5 --result win --killzone NY

# See what the agent has learned
python3 scripts/analyze.py

Journal fields

Field	Value	Description
`symbol`	SP500, BTCUSD…	Instrument
`bias`	LONG / SHORT	Direction
`type`	1, 2, 3	DRT type (continuation/reversal/consolidation)
`entry` / `sl` / `tp`	price	Trade levels
`rr`	1.5, 2.0, 3.0	R:R at entry
`result`	win / loss / be	Outcome
`killzone`	London, NY, SB-AM, SB-PM	Where the trade was taken
`notes`	text	Free note (e.g. "sweep 12 bars old")

Learning logic (analyze.py)

The agent prints concrete learnings, e.g.:

"Type 2 LONG wins 92% — keep taking them"
"Trades in SB-PM lose 60% — avoid or tighten the filter"
"R:R < 1.5 gives 40% WR — skip, wait for 2R+"
"When the sweep is older than 12 bars: 0 winners — set an age gate"

Killzone reminder (proactive)

Use the agent to remind about trading windows (CET/Danish time):

London 09:00-11:00 · NY 14:30-17:00 · SB AM 09:00-10:00 · SB PM 19:30-21:30
⛔ Never NY open 15:30-16:00 · max 3 trades/day · SL ALWAYS

Rules that never change (even if the data says otherwise)

Stop loss ALWAYS · Max 3 trades/day · Never revenge trade
Only A+ setups — a high win rate requires saying no to 80% of setups