Frequently Asked Questions About Lewis Jackson Self-Improving Trading Agent Framework
22 answers covering everything from basics to advanced usage.
// Basics
What is a oneshot prompt in the context of AI trading agents?
A oneshot prompt is a single copy-paste prompt that, when fed into Claude Code, orchestrates the entire trading agent setup end-to-end. It handles environment detection, strategy onboarding, file scaffolding, Railway cloud deployment, and Hermes installation — all without multi-step manual configuration. The prompt is versioned and updated based on community feedback in the 01 Systems community, so always retrieve the latest version rather than copying from a video.
What is a well-defined goal for a trading agent?
A well-defined goal includes both a specific success definition and a specific failure definition. Success might be a 15% monthly return with a Sharpe score ≥ 1.2 and maximum drawdown ≤ 12%. Failure might be two consecutive months below 5% return. The agent uses this polarity — toward-goal vs. away-from-goal — to orient every improvement cycle. Without both poles defined, the self-improvement loop has no direction and the agent is effectively flying blind.
What is a Hermes-readable trade ledger?
A Hermes-readable trade ledger is a structured file containing all historical trades — both wins and losses — converted into a format Hermes can parse, score, and learn from. It is generated automatically during Phase 3 scaffolding. If you have existing trade history, the onboarding agent converts it into this structured format. This ledger is the raw data Hermes uses to analyze patterns, score outcomes, and form improvement hypotheses.
What is the scientific method loop in AI trading?
The scientific method loop is the core self-improvement protocol: change only one variable per iteration, observe the outcome, determine whether it moved toward or away from the defined goal, promote the better-performing version to the new baseline, then form a new hypothesis and iterate. This ensures clean attribution — you know exactly which change caused any performance shift. It prevents the common mistake of changing multiple parameters at once and being unable to identify what worked.
Do I need coding experience to use this framework?
No extensive coding experience is required. The oneshot prompt handles environment detection, scaffolding, file creation, and deployment automatically through Claude Code. You need to be comfortable working in a terminal (pasting commands, navigating to the Claude Code session) and creating accounts on Railway.app. The onboarding flow is guided — you respond to prompts rather than writing code. However, understanding basic concepts like APIs, YAML files, and terminal commands will help you troubleshoot if issues arise.
// How To
How do I connect my existing trading strategy to the self-improving agent?
During Phase 2 of the onboarding flow, choose path A and point the agent at your existing strategy file on your machine by name. The agent will locate, parse, and extract your current parameters — entry/exit rules, position limits, slippage tolerance, and more. It also extracts your trade history and converts it into a Hermes-readable ledger. Review the generated strategy document to confirm it accurately reflects your intent before proceeding.
How do I authenticate Railway for the trading agent?
When the onboarding flow reaches the Railway step, the agent attempts a Railway CLI login. If the interactive login fails inside the Claude Code session, open a split terminal, paste the provided login command, complete browser-based authentication, then return to the original session and type 'done, continuing'. If you don't have a Railway account yet, create one during this step — the free tier is sufficient. This step ensures your agent runs 24/7 regardless of your local machine's state.
How do I monitor the Hermes improvement cycles?
Use the check-in commands provided in the final configuration summary after deployment. Hermes reviews trades on a weekly cadence and produces scored hypotheses in markdown format. Each cycle shows what variable was changed, the outcome measured against your success/failure definitions, and the proposed next change. When a cycle result moves toward your success definition, approve it as the new baseline. Track directional progress over time — toward-goal outputs compound into meaningful strategy improvements.
How do I switch Hermes from read-only to live trading mode?
After reviewing the first Hermes read-only cycle output — which produces a markdown analysis without writing changes to your live strategy — you manually approve the transition by editing the Hermes trading strategy YAML file. Change the operating mode from read-only to live. Do not flip to live mode until you have verified Hermes correctly understood your strategy, goal definitions, and scoring weights. This safety step protects live capital from a misconfigured improvement loop.
// Troubleshooting
What happens if the Railway login fails during setup?
If the Railway CLI interactive login fails inside the Claude Code session, open a second split terminal window, paste the Railway login command provided by the agent, and complete the browser-based authentication in that separate terminal. Once authenticated, return to the original Claude Code session and type 'done, continuing' to proceed. This is a known friction point because some terminal environments don't handle Railway's interactive browser redirect cleanly.
What if my trading agent sets an impossible success target?
The onboarding agent includes a sanity check that flags success definitions mathematically impossible given your starting capital — for example, targeting $1M/month on $10. If flagged, you must revise your success definition to realistic bounds. An impossible target wastes all improvement cycles because the agent cannot converge toward something unreachable. Set targets that are ambitious but physically achievable given your capital, asset volatility, and timeframe.
Why is my self-improving trading agent not getting better over time?
The most common causes are: vague or missing failure definitions (the agent has no polarity to orient improvements), multiple variables being changed simultaneously (corrupted learning signal), inaccurate or inconsistent data feeds (garbage in, garbage out), or an impossible success target the agent cannot converge toward. Check that each cycle changes exactly one variable, your failure threshold is specific and measurable, your data sources are reliable, and your goal is achievable given your capital.
What if the oneshot prompt from the video doesn't work?
Never use the prompt shown in the video recording — it may be outdated. Always retrieve the latest version from the 01 Systems community (Classroom > YouTube Video Prompts). The oneshot prompts are themselves iterated based on community feedback and bug reports. If the latest version still fails, report the issue in the community so it can be patched in the next version. Treating the prompt as static is explicitly listed as a framework pitfall.
// Comparisons
How does this framework compare to using a generic AI chatbot for trading advice?
A generic AI chatbot gives you one-time trading ideas with no feedback loop, no structured goal, and no continuous improvement. This framework deploys a persistent agent that trades autonomously 24/7, tracks every outcome in a structured ledger, and applies the scientific method to iteratively refine its strategy toward your defined success metrics. The chatbot forgets; Hermes compounds learning. The chatbot has no accountability; the framework enforces measurable success and failure definitions.
How does the Lewis Jackson framework compare to building a trading bot from scratch?
Building from scratch requires you to architect the learning loop, data pipeline, deployment infrastructure, and iteration protocol yourself. The Lewis Jackson framework provides all of this via a single oneshot prompt — environment detection, strategy scaffolding, Hermes installation, and Railway deployment are automated. More importantly, the self-improvement loop with single-variable testing and goal polarity is built in. A from-scratch bot typically lacks this structured improvement methodology and requires manual retraining.
Can I use this framework for stocks or forex, not just crypto?
Yes, the framework supports any asset tradeable via API. The target asset is a required input — you can specify a forex pair, stock ticker, or any crypto token. The strategy parameters are configured for your specific asset during Phase 2. The core principles — well-defined goals, single-variable testing, 24/7 reliability, data accuracy — are asset-agnostic. The main constraint is having API access to execute trades on your chosen market.
// Advanced
Can I run multiple self-improving agents for different assets simultaneously?
Yes, each agent is an independent deployment with its own strategy document, goal definitions, Hermes-readable ledger, and Railway instance. Run through the oneshot prompt setup separately for each asset. Each agent's Hermes brain operates on its own weekly cadence and improvement loop. Keep each agent's capital allocation separate and ensure each has independent success/failure definitions. Do not share a single Hermes instance across multiple assets — the learning signals would be conflated.
What is the weekly cadence and why does it matter?
The weekly cadence is Hermes' default review and improvement cycle frequency — once per week, it analyzes all trades since the last cycle, scores outcomes against your goal, forms a hypothesis, and proposes one variable change. The weekly interval provides enough trade data for statistically meaningful analysis while preventing over-optimization from too-frequent changes. If you run a secondary agent like Cornelius, it should be offset by 3 days to prevent simultaneous conflicting parameter updates.
What are score weights and how does Hermes adjust them?
Score weights are configurable parameters in the strategy document that determine how different trade outcomes (win rate, profit factor, drawdown, Sharpe ratio) are weighted relative to your defined goal. Hermes owns these weights and can adjust them as part of its self-improvement process. For example, if your goal prioritizes risk-adjusted returns, Hermes may increase the Sharpe score weight and decrease the raw return weight. These adjustments are single-variable changes subject to the scientific method loop.
How do I ensure data accuracy for my trading agent?
Establish strict, rules-based interpretation criteria for all data inputs before deploying. Use consistent API sources — don't mix data from different providers with different formats. For non-numerical inputs like news sentiment, define objective parsing rules rather than relying on the AI's interpretation, which can drift. Unreliable API connections or ambiguous data parsing corrupt every downstream decision. Answer 'How will I ensure data accuracy?' in writing before starting setup — this is Step 1 of the workflow.
What does the first read-only Hermes cycle look like?
The first Hermes cycle is observation-only. Hermes reviews your existing trade history and current strategy, scores outcomes against your success and failure definitions, and produces a markdown report. This report includes its understanding of your strategy, its scoring of historical trades, and preliminary hypotheses — but it does not write any changes to the live strategy. You review this output to verify Hermes correctly interpreted your goals and parameters before manually approving the transition to live write mode.
Can I manually override Hermes' strategy changes?
Yes. You approve strategy promotions — Hermes proposes changes, but you decide whether a winning cycle's output becomes the new baseline. You can also edit the Hermes trading strategy YAML directly to override parameters. However, if you change multiple variables manually between cycles, you corrupt the single-variable learning signal. If you want to make a manual override, treat it as the sole variable change for that cycle and let Hermes evaluate the result in the next review.