Gaurang1745

AI Trading Bot — Data skill for Claude Code

Data community

Autonomous AI trading experiment on Indian equities (NSE/BSE), driven by Claude Opus 4.7 + a multi-agent pipeline.

How to install AI Trading Bot

This entry records only its repository, not the path inside it, so there is no exact command to give. Open Gaurang1745/ai-trading-bot and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What AI Trading Bot does

Autonomous AI trading experiment on Indian equities (NSE/BSE), driven by Claude Opus 4.7 + a multi-agent pipeline. 14-session live run, full archive on Vercel.

Alternatives in Data

Autonovel — An autonomous novel writing pipeline, by Hermes Agent 1.5k ★
Claude Code Memory Setup — Up to 71.5x fewer tokens per session on Claude Code with Obsidian + Graphify 947 ★
Agentsview — Local-first session intelligence and analytics for coding agents, supporting Claude Code, Codex, and 14 other 824 ★

README

AI Trading Bot

An autonomous equity-trading experiment on Indian markets (NSE/BSE), with **Claude Opus 4.7** making every trading decision and a Next.js dashboard exposing every trade, prompt, and decision the bot made.

**Live archive dashboard:** **https://ai-trading-bot-archive.vercel.app**

The experiment ran 2026-04-22 → 2026-05-11 (14 trading sessions) and is now concluded. The dashboard is frozen on the final state and serves the full archive — `/` is a project showcase, `/dashboard` has every trade and snapshot, `/logs` lets you browse every Sonnet, Opus, and Haiku call (system prompt, user prompt, response, metadata).

The Experiment

The question we're trying to answer: **how well does Claude Opus 4.7 perform as an autonomous discretionary trader if we give it good market data, deep stock research, and a structured decision loop — and stay out of its way?**

**Setup**


Capital	₹10,00,000 (paper)
Universe	NSE/BSE — large + mid caps + select ETFs (~510 instruments)
Started	2026-04-22
Mode	Paper trading via realistic OHLC fill simulation against live Dhan market data
Decision cadence	Every 30 min during market hours (12 cycles/day)
Models	Sonnet 4.6 (market pulse) → Opus 4.7 (deep research + trading decisions) → Haiku 4.5 (news summarization)
Risk gates	15+ deterministic guardrails between Claude's decision and order placement

The bot is mode-blind: Claude never knows it's paper trading. The data format, prompts, and execution interfaces would be identical in a live-trading run.

Architecture

                         ┌──────────────────────────┐
                         │   Dhan HQ (broker API)   │
                         │  market data + LTPs      │
                         └────────────┬─────────────┘
                                      │
   ┌──────────────────────────────────▼──────────────────────────────┐
   │  ai-trading-bot/  (