Skip to content

docs: add AGENTS.md agent operating guidance - #227

Open
andrewluxem wants to merge 1 commit into
TorchTrade:mainfrom
andrewluxem:docs/agent-operating-kit
Open

andrewluxem wants to merge 1 commit into
TorchTrade:mainfrom
andrewluxem:docs/agent-operating-kit

Conversation

@andrewluxem

Copy link
Copy Markdown

Summary

  • Adds a self-contained, vendor-neutral AGENTS.md for TorchTrade.
  • Keeps the tracked change to one docs-only file.
  • Respects the existing repo convention that CLAUDE.md and .claude/ are git-ignored local maintainer tooling.

Repo tailoring

  • Treats TorchTrade as a shared, multi-maintainer OSS RL-for-trading library.

  • Adds explicit financial/trading safety boundaries:

    • no live trading
    • no order placement
    • no broker/exchange or account mutations
    • no portfolio changes
    • no real-money execution
    • no investment advice, trading recommendations, guaranteed-return language, or performance claims
  • Names protected risk surfaces:

    • broker/exchange credentials
    • .env and credential variables
    • datasets and generated outputs
    • model artifacts and checkpoints
    • notebooks
    • trading logs
    • live and paper trading adapters
    • external-service integrations such as wandb, Hugging Face, OpenAI/vLLM/LLM calls
  • Documents approval-gated commands and actions:

    • uv sync
    • uv run pytest
    • mkdocs serve
    • model training
    • backtests
    • data downloads
    • live/paper broker or exchange calls
  • Notes the existing @claude GitHub Action reads AGENTS.md, so AGENTS.md is the appropriate vendor-neutral shared agent guidance file.

Validation

  • Confirmed only AGENTS.md was committed.
  • Confirmed CLAUDE.md and .claude/ remain ignored/untracked and were not force-added.
  • Confirmed no source, tests, examples, package/config files, workflows, datasets, model artifacts, notebooks, generated outputs, trading logs, runtime files, or secrets changed.
  • Confirmed git diff --cached --check was clean before commit.
  • Confirmed no real credential values were introduced; policy words and placeholder variable names only.

Safety

  • Docs-only change.
  • No install, test, training, backtest, data download, broker/exchange call, order placement, live trading, external API, release, publish, deploy, or automation action was run.
  • No secrets, .env values, broker/exchange keys, account balances, portfolio data, private datasets, generated private outputs, or trading logs were inspected or printed.
  • No financial advice, performance claim, guaranteed-return language, trading recommendation, or investment recommendation was added.

Add a self-contained, vendor-neutral AGENTS.md so AI coding agents such as Claude Code, Codex, Cursor, and the @claude GitHub Action operate within clear safety boundaries for this RL-for-trading library.

The guidance forbids live trading, order placement, broker/exchange or account mutations, secret handling, financial advice, performance claims, and guaranteed-return language. It documents the stack, protected surfaces, approval-gated commands, validation expectations, and the shared-upstream branch/contribution model.

The repo intentionally git-ignores CLAUDE.md and .claude/, so those local maintainer files are not committed and the existing .gitignore convention is preserved.

Docs-only. No source, tests, examples, package/config files, workflows, datasets, model artifacts, notebooks, generated outputs, trading logs, runtime files, or secrets changed. No install, test, training, backtest, data download, broker/exchange call, order placement, live trading, external API, release, publish, deploy, or automation action was run.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant