by Blave-TW
Quant infrastructure for AI agents — a free, open-source desktop workspace for macOS and Windows where your Claude Code or Codex turns a trading idea into a backtested strategy and runs it live.
# Add to your Claude Code skills
git clone https://github.com/Blave-TW/blave-agentLast scanned: 10/23/2026
{
"issues": [],
"status": "PASSED",
"scannedAt": "2026-10-23T09:08:09.027Z",
"npmAuditRan": true,
"pipAuditRan": true,
"promptInjectionRan": true
}See how blave-agent compares with popular alternatives.
blave-agent is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by Blave-TW. Quant infrastructure for AI agents — a free, open-source desktop workspace for macOS and Windows where your Claude Code or Codex turns a trading idea into a backtested strategy and runs it live. It has 152 GitHub stars.
Yes. blave-agent passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.
Clone the repository with "git clone https://github.com/Blave-TW/blave-agent" and add it to your Claude Code skills directory (see the Installation section above).
blave-agent is primarily written in Python. It is open-source under Blave-TW on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh blave-agent against similar tools.
No comments yet. Be the first to share your thoughts!
30 days in the Featured rail · terms & refunds
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
Automated security scan — not a certification.
Build, backtest and trade strategies live with Claude Code or Codex. Free and open source.
English | 繁體中文 | 简体中文 | 日本語 | Español | Português | Tiếng Việt
https://github.com/user-attachments/assets/7b33edb7-9c65-4e19-854a-40295c6e8b74
Quick start · Run it with your computer off
Star the repo if this is useful — and Watch › Releases to get notified of new versions.
Claude Code or Codex can write a trading bot. Running it for real is the hard part. In a terminal, every strategy is a folder of scripts and a chat log, and past a handful they are hard to keep track of. Wiring up your own auto-trading means handling scheduling, exchange APIs, position reconciliation, restarts and keys yourself. Blave Agent is that missing part.

Each strategy sits in the sidebar with its backtest, trades, parameter scan, out-of-sample test, code and saved versions. Restore an old version from the list, and set how much each strategy trades on the Portfolio page.
state/audit.jsonl).Overfitting: parameters that just happen to fit past data.
lib/validation.py): could shuffled data have done as well?lib/param_scan.py); walk-forward for out-of-sample results (lib/walk_forward.py).lib/quality_check.py).AGENTS.md).
Press Search the Web for Ideas and the agent reads a dozen-odd sites, then brings back three backtestable ideas, each with entry and exit rules and its source URL. Pick one and it builds and backtests it. The pages it reads open in the app's built-in browser, on your screen; exchange account pages and private-network addresses are blocked.
Download the app for macOS or Windows (macOS 13+, or Windows 10/11 x64) and open it.
On first launch you choose what powers the agent:
Then describe an idea:
The agent sorts every idea into one of three types before writing code:
| Type | What it is | Backtest |
|---|---|---|
| A | One fixed symbol on a fixed interval; one position (long / short / flat) | Required |
| C | A portfolio: N symbols and a weight vector that sums to at most 1, rebalanced on a schedule | Required |
| B | Everything else: screeners, grids, arbitrage, alerts, one-off execution | None |
shell/package.json › engines. The packaged app bundles its own Python 3.12; running from source uses your system Python (python3, or python on Windows) to create the venv.winget install --id 9PLM9XGG6VKS -s msstore), ChatGPT.app on Mac. On Windows, avoid the npm @openai/codex 0.161–0.162.1 builds for now (their sandbox is broken).npm.cmd runs even when PowerShell's execution policy blocks the npm script.BLAVE_LANG=en npm start (PowerShell: $env:BLAVE_LANG="en"; npm.cmd start).You need macOS 13+ or Windows 10/11 (x64), Node.js 22.12+, Python 3 on your PATH, and one of: Claude Code or Codex signed in, a DeepSeek API key, or a Blave account.
git clone https://github.com/Blave-TW/blave-agent.git
cd blave-agent/shell
npm install
npm start
On Windows, run npm.cmd install and npm.cmd start in PowerShell.
What changed in each version: Releases.
Venues with a tested order library (lib/account_*.py + lib/order_*.py, verified on real accounts):
references/capital-broker.md) — the desktop app cannot connect it on any OS, a Windows PC included. You upload your certificate in the app and the agent completes the setup on the cloud machine — no remote desktop needed.references/president-broker.md). On the Windows desktop app you upload the certificate and sign in inside the app. On a Windows cloud machine you apply for or export the certificate on your own Windows PC, upload it on the web workspace's Auto trading page and confirm the sign-in there — no remote desktop.How a venue gets connected depends on where the agent runs:
lib.venue.bind.For any other exchange or broker with an API, the agent can write a helper from lib/account_TEMPLATE.py and lib/order_TEMPLATE.py. That code has not been tested by Blave.
Paper trading needs no keys. It ships as lib/account_paper.py + lib/order_paper.py and fills at the market's latest real price, never an adjusted one: crypto at Binance's last trade, TAIFEX futures at the latest minute, stocks at the unadjusted daily close (lib/paper_data.py).
Data.
lib/data.py — crypto klines and indicators, Taiwan stocks, futures. US stock and ETF daily bars are fetched the same way as on the desktop app.