How to Systematize Your Crypto Strategy With a Self-Improving Agent
For Discretionary crypto traders · Based on Lewis Jackson Self-Improving Trading Agent Framework
// TL;DR
Discretionary crypto traders often have a working strategy but no way to systematically improve it. The Lewis Jackson Self-Improving Trading Agent Framework lets you feed your existing strategy file and trade history into Claude Code via a oneshot prompt, which scaffolds a Hermes-readable ledger and deploys a 24/7 agent on Railway. Hermes then reviews your trades weekly, forms scored hypotheses, and changes one variable per cycle to compound improvements toward a defined return and Sharpe goal. Use it when you're tired of eyeballing what to tweak and want disciplined, attributable strategy evolution running around the clock.
Why can't I improve my crypto strategy consistently?
Most discretionary crypto traders hit the same wall: you have a strategy that works sometimes, weeks of trade history, and a gut feeling about what to fix — but no clean way to know which change actually helped. You tweak position sizing, entry timing, and stop-losses all at once, results move, and you can't attribute the change to anything. The Lewis Jackson Self-Improving Trading Agent Framework replaces that guesswork with a disciplined scientific loop that changes one variable at a time.
How do I get my existing strategy into the agent?
Start by writing your four criteria: how you'll ensure data accuracy, how you'll guarantee 24/7 reliability, your exact success definition (say, 15% monthly return with Sharpe ≥ 1.2 and max drawdown ≤ 12%), and your exact failure definition (say, two consecutive months below 5% return). Then pull the latest oneshot prompt from Lewis Jackson's 01 Systems community and paste it into Claude Code.
In Phase 2, choose path A and point the agent at your existing strategy file by name. It parses your position limits, entry signals, and slippage tolerance automatically. Phase 3 converts your trade history — even 50 mixed wins and losses across six weeks — into a Hermes-readable ledger. This becomes the raw material Hermes analyzes.
What will Hermes actually change first?
Hermes runs read-only for the first week. It might discover, for example, that your position sizing on losing trades was 40% larger than on winning trades. That's one variable. It proposes reducing max position size by 20% as the sole change for the next cycle. You review the markdown analysis, confirm Hermes understood your strategy, then approve. If the next cycle moves toward your goal, that version becomes the new baseline.
The discipline matters. Because only one variable changes per cycle, every improvement is traceable and durable. Over multiple weekly cycles, these clean gains compound instead of tangling into noise.
How do I keep it running while I sleep?
Crypto never closes, so a local-only agent that dies when your laptop sleeps is useless. During setup, you authenticate with Railway for 24/7 cloud hosting. If the CLI login fails inside the session, open a split terminal, run the provided login command, complete browser auth, and type 'done, continuing.' Railway is free up to a significant threshold and pushes strategy updates automatically.
What should I avoid?
Don't flip to live mode before reviewing the first read-only cycle — that review confirms Hermes hasn't misread your strategy before it touches real capital. Don't set an impossible goal relative to your capital; the agent will flag targets like 10x in 30 days on minimal funds. And never modify multiple variables between cycles yourself — it corrupts the learning signal Hermes depends on. Finally, if you parse news or sentiment, set objective interpretation rules so conclusions don't drift with the AI's interpretation.
What next step should I take?
Write your four criteria today — success definition, failure definition, data accuracy plan, and reliability plan. Then grab the latest oneshot prompt from the 01 Systems community, open Claude Code, and run Phase 1. Within a session you'll have a scaffolded, Railway-hosted agent producing its first read-only Hermes review of your real trade history.
// FREQUENTLY ASKED QUESTIONS
Can I use my past crypto trades even if most were losses?
Yes — Hermes learns from both wins and losses. During Phase 3, your full trade history is converted into a Hermes-readable ledger. Losing trades are especially valuable because they often reveal the clearest patterns, like oversized positions on losers, that become the first single-variable change Hermes proposes.
Which crypto assets can the agent trade?
You specify the target asset during setup — Bitcoin, Ethereum, Solana, a specific subnet token, or another market. The agent scaffolds position limits, slippage tolerance, and gas reserve appropriate to that asset. You can run separate agents per asset, but offset their review cadences to avoid conflicting parameter updates.
How often will my strategy actually change?
Hermes reviews on a weekly cadence and changes at most one variable per cycle. So expect roughly one tested change per week, with winning versions promoted to the new baseline. This deliberate pace keeps every improvement attributable and compounding rather than fast but untraceable.