Algo-Trading-Skills

A structured algorithmic trading skills library for AI coding agents

28 production-grade algorithmic trading skills built · 502 skills tracked across a global research roadmap spanning 16 engineering domains · broker/exchange coverage spanning India (Fyers, Zerodha Kite, ICICI Breeze, Upstox), US (Alpaca, IBKR, Schwab, TradeStation), global crypto (Binance, Coinbase, Kraken, Deribit), forex (OANDA, MT5), and dozens more venues in the roadmap · agentskills.io standard

Status note: this repo is mid-expansion from an India-first initial pass to global coverage. 28 skills currently have the full SKILL.md + references/ + scripts/ + assets/ structure and pass tools/validate_skills.py. A further 474 are tracked as titled, scoped entries in docs/ROADMAP_500.md — a prioritized research backlog covering global exchanges, regulatory regimes, execution algorithms, custody, and more, not finished work. Broker/regulatory specifics in planned entries should be verified against current sources before being built out into a real skill. See Contributing if you want to help.

Get Started · What's Inside · Skill Anatomy · Platforms · Contributing

---

Community Project. This is an independent, community-created skills library. Not affiliated with Anthropic PBC or any broker named in this repo.

Engineering guidance, not financial advice. These skills encode production engineering practices for trading infrastructure. They do not guarantee the profitability of any strategy and do not eliminate the risk of loss in live trading. See the Disclaimer.

Give any AI agent the instincts of a senior trading-infrastructure engineer

An AI coding agent can write a WebSocket client, a backtest loop, or an order-placement function that looks completely correct — right library calls, clean structure, plausible logic — and still fail in production for reasons that have nothing to do with code quality: a broker invalidates a token overnight in a way its docs don't mention, a backtest silently uses a bar's own close to predict its own direction, a risk limit lives inside the same function it's supposed to constrain, a WebSocket callback blocks the read loop during exactly the volatility spike a strategy exists to catch.

Your AI agent doesn't know these failure modes — unless you give it these skills.

This repo contains 18 structured skills spanning 6 engineering domains of algorithmic trading infrastructure, each following the agentskills.io open standard, drawn from a real, running Nifty 50 options trading platform: dual-broker authentication (Fyers REST + ICICI Breeze via Selenium), an ML signal classifier with walk-forward validation, a producer-consumer tick pipeline with explicit backpressure policy, correlation-aware position sizing, and systemd-supervised deployment.

Why this exists

Cybersecurity has Anthropic-Cybersecurity-Skills, an 800+ skill library mapped to MITRE ATT&CK and NIST, giving AI agents the structured decision-making a senior security analyst follows. Algorithmic trading has had no comparable resource — existing repos give you broker SDKs, indicator libraries, or strategy templates, but none give an agent the practitioner playbook for when to use a technique, what to check first, how to execute it step by step, and how to verify it actually worked.

This is a first pass at that resource for trading: not hundreds of skills, but 18 that are each deep enough to prevent a specific, named class of production bug. Quality over volume — every skill here answers yes to: would this have actually prevented a real production bug, and is it specific enough for an agent to follow step-by-step rather than nod along with generic advice?

What's inside — 16 categories

The first 6 categories are the original India-first pass; the next 10 extend the same quality bar to global markets — crypto exchanges, forex brokers, multi-currency and multi-timezone data handling, non-Indian regulatory regimes, multi-asset derivatives, execution algorithms, custody/security, cross-strategy portfolio management, market microstructure, alternative-data research, and tax/accounting.

Domain	Built	Tracked (built+planned)	Key coverage
broker-integration	4	36	Headless auth (REST + Selenium), token lifecycle via live probing, order idempotency, per-broker rate limiting
real-time-architecture	4	30	Producer-consumer tick pipelines, burst-safe buffering, explicit backpressure policy, WebSocket reconnection without duplicate subscriptions
backtesting-methodology	3	30	Lookahead bias elimination, walk-forward validation, realistic slippage/fee/latency simulation
financial-ml	3	38	Leakage-free feature engineering, offline-train/online-infer deployment without train/serve skew, live model staleness detection
risk-management	2	39	Kill switches and drawdown circuit breakers, correlation-aware exposure limits
deployment-ops	2	30	systemd process supervision, the paper-to-live promotion gate
global-market-integration	2	44	Crypto exchange APIs (Binance/Coinbase/Kraken), forex brokers (OANDA/MT5), dozens of global venues in roadmap
regulatory-compliance-global	2	38	US Pattern Day Trader rule, EU MiFID II/RTS 6, and a global regulatory roadmap (CFTC, IIROC, SFC, FSA, SEBI, ASIC, MAS, FINMA...)
multi-asset-derivatives	1	28	Options/futures SPAN-style margin estimation
execution-algorithms	1	33	TWAP/VWAP order slicing
data-management-global	3	37	Global exchange holiday calendars, multi-timezone/DST-safe scheduling, multi-currency P&L
crypto-custody-security	1	29	Crypto wallet/API-key custody and permission scoping
portfolio-multi-strategy	0	30	(planned — see roadmap)
market-microstructure-latency	0	24	(planned — see roadmap)
quant-research-alt-data	0	20	(planned — see roadmap)
tax-accounting-reporting-global	0	16	(planned — see roadmap)

Full indexed list with build status: index.json. Full 502-entry roadmap with one-line scope for every planned skill: docs/ROADMAP_500.md.

The 28 built skills

Skill	Category
headless-broker-auth-patterns	broker-integration
token-lifecycle-live-probing	broker-integration
order-placement-idempotency	broker-integration
multi-broker-rate-limit-handling	broker-integration
producer-consumer-tick-pipeline	real-time-architecture
tick-buffering-burst-handling	real-time-architecture
backpressure-drop-degrade-policy	real-time-architecture
websocket-reconnect-without-duplicate-subscriptions	real-time-architecture
lookahead-bias-elimination	backtesting-methodology
walk-forward-validation-setup	backtesting-methodology
execution-realistic-simulation	backtesting-methodology
feature-engineering-without-leakage	financial-ml
offline-train-online-infer-deployment	financial-ml
model-staleness-detection	financial-ml
kill-switch-and-drawdown-circuit-breakers	risk-management
correlation-aware-exposure-limits	risk-management
systemd-supervision-for-trading-bots	deployment-ops
paper-to-live-promotion-checklist	deployment-ops
crypto-exchange-api-integration	global-market-integration
forex-broker-integration-oanda-mt5	global-market-integration
pattern-day-trader-rule-compliance-us	regulatory-compliance-global
mifid-ii-algo-trading-compliance-eu	regulatory-compliance-global
options-margin-span-calculation-global	multi-asset-derivatives
execution-algo-twap-vwap-slicing	execution-algorithms
global-exchange-holiday-calendar-handling	data-management-global
multi-timezone-session-scheduling	data-management-global
multi-currency-pnl-and-fx-conversion	data-management-global
crypto-wallet-key-custody-security	crypto-custody-security

See docs/architecture.md for how these fit together as a system, and mappings/broker-api-coverage.md / mappings/regulatory-coverage.md for cross-cutting broker and regulatory touchpoints.

Quick start

git clone https://github.com/<your-org>/algo-trading-skills.git
cd algo-trading-skills
python tools/validate_skills.py   # optional: confirms every skill's structure/frontmatter

Point your agent at the skills/ directory (see Compatible platforms below for the exact wiring per tool), and it can discover and load skills by scanning SKILL.md frontmatter.

How AI agents use these skills

Each skill costs roughly 30–50 tokens to scan (frontmatter only) and 500–1,500 tokens to fully load (complete workflow in SKILL.md, more in references/ if needed). This progressive-disclosure structure — mirrored from the anatomy below — lets an agent search all 28 built skills (and 502 tracked overall) without blowing its context window.

User prompt: "My Fyers bot's live orders keep getting placed twice after a timeout"

Agent's internal process:

  1. Scans 18 skill frontmatters (~30-50 tokens each)
     → identifies order-placement-idempotency and token-lifecycle-live-probing
       as the relevant matches

  2. Loads skills/order-placement-idempotency/SKILL.md in full
     → follows the Workflow section: classify timeout as ambiguous (not
       failed), reconcile against the broker order book before any retry

  3. Loads references/workflows.md for the full procedure and
     scripts/order_ledger.py for a working starting point

  4. Validates the fix using the Verification section
     → confirms a simulated timeout no longer produces a duplicate order

Skill anatomy

Every skill follows a consistent directory structure:

skills/order-placement-idempotency/
├── SKILL.md              ← Skill definition (YAML frontmatter + Markdown body)
├── references/
│   ├── standards.md      ← Broker/framework coverage + regulatory touchpoints
│   └── workflows.md      ← Deep technical procedure reference
├── scripts/
│   └── order_ledger.py   ← Working helper script
└── assets/
    └── checklist.md      ← Sign-off checklist

Full anatomy and frontmatter field reference: docs/skill-anatomy.md.

YAML frontmatter (real example)

---
name: order-placement-idempotency
description: >-
  Use whenever a bot places, modifies, or cancels live orders and must
  guarantee it never double-executes an order due to retries, timeouts,
  or reconnects
domain: algorithmic-trading
subdomain: broker-integration
tags: ["broker-integration", "fyers-api-v3", "zerodha-kite-connect", "icici-breeze-api"]
brokers_frameworks: ["Fyers API v3", "Zerodha Kite Connect", "ICICI Breeze API", "Upstox API v2", "Alpaca Trading API", "IBKR API"]
version: "1.0"
author: algo-trading-skills-contributors
license: Apache-2.0
---

Markdown body sections

## When to Use          Trigger conditions — when should an AI agent activate this skill?
## Prerequisites        Required tools, access, and environment setup.
## Workflow             Step-by-step execution guide with specific decision points.
## Common Pitfalls      Named, specific failure modes this skill prevents.
## Verification         How to confirm the skill was executed successfully.
## Related Skills       Cross-links to other skills in this repo.

tools/validate_skills.py enforces this structure in CI (see .github/workflows/validate-skills.yml).

Compatible platforms

AI coding agents Claude Code · Cursor · GitHub Copilot · OpenAI Codex CLI · Gemini CLI · Windsurf · Cline · Continue

Agent frameworks Any framework that can read plain Markdown + YAML frontmatter from a filesystem path (LangChain, CrewAI, custom MCP-based agents, etc.)

Installation differs slightly by platform:

• Claude Code — copy or symlink skills/* into .claude/skills/ (project-level) or ~/.claude/skills/ (global), or install via the .claude-plugin/plugin.json manifest if your setup supports plugin-style discovery.
• Cursor — point Cursor's rules/skills configuration at the cloned skills/ directory.
• GitHub Copilot — reference skill files as context, or copy into .github/copilot-instructions/ if supported.
• Gemini CLI / other agentskills.io-compatible tools — load directly from a cloned copy of this repo per your tool's skills-directory convention.

The format is deliberately plain (YAML frontmatter + Markdown, no platform-specific syntax) so manual copy-paste into any agent's context window always works as a fallback.

Architecture context

The skills assume (and are easiest to apply within) a system shaped like a producer-consumer tick pipeline feeding a strategy/ML engine, gated by an independent risk module, behind idempotent order placement, supervised by systemd — see docs/architecture.md for the full diagram and how each skill maps onto it.

Contributing

This project grows through contributions that reflect genuine production experience. See CONTRIBUTING.md for the process, and skills/ for domains that could use expansion — this first pass deliberately covers 18 deep skills rather than attempting broad coverage; areas like options-Greeks-specific risk modeling, multi-asset portfolio construction, and crypto-exchange-specific integration are open for contribution.

Every PR is reviewed against the quality bar in CONTRIBUTING.md and the structural checks in tools/validate_skills.py.

Community

• Issues — bug reports and skill proposals (templates provided)
• Pull requests — see CONTRIBUTING.md before opening
• Security policy — responsible disclosure for guidance or script vulnerabilities

Citation

If you use this project in research or documentation, see CITATION.cff.

License

Licensed under the Apache License 2.0 — free to use, modify, and distribute in personal and commercial projects.

Disclaimer

These skills encode engineering practices for building trading infrastructure. They are not financial advice, do not guarantee the profitability of any strategy, and following them does not eliminate the risk of loss in live trading. Always validate thoroughly in paper trading (see skills/paper-to-live-promotion-checklist/SKILL.md) before committing real capital, and confirm the regulatory requirements applicable to algorithmic trading in your jurisdiction independently (see mappings/regulatory-coverage.md).

---

Community project. Not affiliated with Anthropic PBC or any broker referenced in this repository.