agentleFS
F0rger123F0rger123/ultimate-claude-trading-skill/AGENTS.md

ultimate-claude-trading-skill

AGENTS.md1 starsChanged 8 days agoMITView on GitHub

If someone pasted this repo's link into Claude, ChatGPT or another assistant, read this first. It tells you how to walk them through the project safely. The project is honest about results. As of 2026-11-09, every tested candidate was positive out of sample but none is statistically proven; the best are paper-tested. It is not financial advice, and most people who trade with leverage lose money. It worked if the tests say OK and reports/corrected/summary.md exists. Read that report to…

The file

We split this AGENTS.md into the rules it gives an agent

8 rules. About 1 is common, 6 are distinctive, and 1 is very specific to this project.

  • ≈1

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    If someone pasted this repo's link into Claude, ChatGPT or another assistant, read this first.

    line 3
  2. Distinctive:

    Only if they ask about real money: "Do you have a verified Coinbase account with futures turned on?"

    line 20
  3. Common:

    Never ask for, print or store API keys or secrets in chat, files or logs.

    line 50
  4. Very specific:

    Never place a real order, set TRADEDESK_LIVE=1, or unlock live trading for them.

    line 51
  5. Distinctive:

    Never spend money or move funds.

    line 52
  6. Distinctive:

    Never lower a test bar so a strategy passes, or describe backtest results as a promise.

    line 53
  7. Distinctive:

    Never help someone use an exchange their country is barred from, for example with a VPN.

    line 54
  8. Distinctive:

    Never touch a live account or "production" without the person's clear yes for that exact step.

    line 55

≈ “Common” is an estimate: at least 1 in 5 of the 42,283 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
AGENTS.md

AI agent guide

If someone pasted this repo's link into Claude, ChatGPT or another assistant, read this first. It tells you how to walk them through the project safely.

1. What this is, in plain words

A trading "desk" that runs on the person's own computer:

  • It tests trading strategies on years of past crypto prices, with fees counted (a backtest).
  • It keeps testing the ones that pass, first on new data and then with pretend money (forward test, then paper trading).
  • It can trade real money on Coinbase's US crypto futures only after the person unlocks it, and it always uses safety limits.
  • It learns from its own trades, but a change is only used after it proves itself on data it has never seen.

The project is honest about results. As of 2026-11-09, every tested candidate was positive out of sample but none is statistically proven; the best are paper-tested. It is not financial advice, and most people who trade with leverage lose money.

2. Ask the person first

  1. "Which computer will run this, and is it on all the time?" Crypto trades around the clock.
  2. "Do you have Python 3.10 or newer?" If they aren't sure, help them check: python --version.
  3. "Do you just want to see the test results, or also run paper trading?"
  4. "Do you have your own strategy or TradingView indicator you want tested?" If so, get the indicator name and settings.
  5. Only if they ask about real money: "Do you have a verified Coinbase account with futures turned on?"

3. Set it up and check it worked

git clone https://github.com/F0rger123/ultimate-claude-trading-skill
cd ultimate-claude-trading-skill
python -m unittest discover -s tests      # expect: OK
python -m tradedesk data --gran 3600 --days 2870   # hourly history since 2019, free public data, no account
python research/corrected_tournament.py            # about 2 minutes
python research/summarize_corrected.py             # writes reports/corrected/summary.md
python -m tradedesk status

It worked if the tests say OK and reports/corrected/summary.md exists. Read that report to them in plain words: what passed, what failed, and why.

In Claude Code you can also install the skills as a plugin:

/plugin marketplace add F0rger123/ultimate-claude-trading-skill
/plugin install ultimate-claude-trading-skill@ultimate-claude-trading-skill

4. How to use it

  • Paper trading: python -m tradedesk desk keeps running and journals trades. Strategies only trade once they reach the forward stage.
  • Phone dashboard API: python -m tradedesk serve --host <tailscale-ip>. It needs TRADEDESK_API_TOKEN set as an environment variable on the PC.
  • Learning loop: python -m tradedesk learn.
  • Weekly shake-up: the desk retires the 3 worst paper accounts every Monday and starts 3 new ones (paper only). Preview with python -m tradedesk rotate --dry-run.
  • Rules and kill switch: in state/rules.json or the phone app.
  • The skills in skills/ hold the details for each task. docs/BLUEPRINT.md is the source of truth.

5. Never

  • Never ask for, print or store API keys or secrets in chat, files or logs. Keys go only in the person's own environment variables, typed by them.
  • Never place a real order, set TRADEDESK_LIVE=1, or unlock live trading for them. Those are their steps.
  • Never spend money or move funds.
  • Never lower a test bar so a strategy passes, or describe backtest results as a promise.
  • Never help someone use an exchange their country is barred from, for example with a VPN.
  • Never touch a live account or "production" without the person's clear yes for that exact step.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.