This repository is a plain Python application for the FourMeme collector, dataset, hybrid training, and bot workflow.
- Read root
AGENTS.mdbefore making changes. - Treat
AGENTS.mdas the canonical repo guidance;CLAUDE.mdmirrors practical workflow notes for Claude Code. - If you edit inside
config/,src/,src/core/,src/data/,src/trader/, ortools/, also read the nearest childAGENTS.md. - Do not edit
docs/goals/unless the current task explicitly asks for goal-document changes. - Treat real trading behavior as high risk. Keep
ENABLE_TRADING=falseunless the task explicitly covers live trading.
python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .envFill .env with local RPC and wallet values. Never commit .env, private keys, wallet files, or paid RPC credentials.
Use unittest commands:
python -m unittest discover
python -m unittest tests.core.test_rpc_config
python -m unittest tests.model.test_run_hybrid_training_cli
python -m unittest tests.smoke.test_surviving_workflow_importsFor narrow changes, run the most relevant targeted test first, then run broader coverage when the change affects shared behavior.
- Generated lifecycle data, datasets, logs, pid files, local caches, and virtualenvs stay untracked.
- Curated replay and model JSON/JSONL reports are reviewable evidence and may be tracked.
- Model binaries under
data/models/are ignored by default unless a reviewed run is intentionally force-added. - Do not change
.env.examplewithout also updating config contract tests when behavior changes.
- Tests or an explicit reason they could not be run.
git diff --checkpasses.git status --short --untracked-files=allcontains only intended changes.- No secrets, wallet material, RPC credentials, or accidental large artifacts.
- Any config-sensitive runtime change updates
.env.exampleand relevant docs/tests.