Skip to content

v3.0.0501 skillsagentskills.ioApache-2.0

Every skill declares
where it stops.

501 playbooks for AI agents that write trading systems, each with a working reference implementation and a test suite. Every one names the situations it covers, the ones it does not, and the skill that takes over. That graph of handoffs is drawn behind you, from the library itself.

Browse the catalog

Skills

501

Domains

16

Handoffs

3,317

Reference implementations

501

Why a library and not a prompt

Code that reads
as correct.

None of these are syntax errors, and none of them look wrong in review. They are the assumptions a trading system punishes. Each one is owned by a skill.

  1. 01

    Broker Integration

    The request timed out, so the order failed. Retry it.

    A timeout means unknown, never failed. Reconcile against the broker's order book before any retry, or the position doubles.

    order-placement-idempotency
  2. 02

    Backtesting Methodology

    Use the bar's close to decide whether to enter on that bar.

    The close is not knowable until the bar completes. A backtest that reads it is predicting the present.

    lookahead-bias-elimination
  3. 03

    Risk Management

    Check the drawdown limit at the top of the strategy function.

    A limit inside the function it constrains can be skipped by the same bug it exists to catch. It needs veto power from outside.

    kill-switch-and-drawdown-circuit-breakers
  4. 04

    Real-Time Architecture

    On reconnect, resubscribe to every symbol.

    The server may still hold the old subscriptions. Reconcile what is actually subscribed, or every tick arrives twice.

    websocket-subscription-reconciliation-after-reconnect

Every skill, the same shape

Seven sections.
501 times.

The structure is not a convention contributors are asked to follow. A validator in CI rejects a skill missing any heading, any file, or naming a skill that does not exist.

01When to Use
The trigger, stated as a situation. Every description opens with “Use when…”, so an agent matches on the symptom in front of it rather than the subject it guesses.
02When NOT to Use
Where the skill stops. Each excluded case names the skill that owns it. This one section is why the library is a graph, and why a near-miss match is handed on instead of applied.
03Prerequisites
What must already be true before the workflow is safe: a broker field that echoes back, a durable ledger, a clock you trust.
04Workflow
The procedure, in order. The reference implementation in scripts/ follows it step for step, so the prose and the code cannot disagree without a test noticing.
05Common Pitfalls
The named failure modes: the ones that pass review and fail in production. Would following this skill have prevented a real incident is the bar for inclusion.
06Verification
A command you can run. Every skill quotes its own test suite, and the sign-off checklist in assets/ is derived from this section.
07Related Skills
Neighbours in the graph, each with the reason it is adjacent. Every slug named here must exist, or the build that validates the library fails.
Six files, machine-enforced

SKILL.md

references/standards.md

references/workflows.md

scripts/<helper>.py

scripts/test_<helper>.py

assets/checklist.md

The scale

2.0M words.
Read one at a time.

About 0.8 million words of SKILL.md and another 1.2 million in references and checklists, beside 521,000 lines of Python. No context window holds it, and none needs to. An agent reads the index, opens one document, and reads its boundary before anything else.

Every skill, by length, in ring order. Hover for the slug.

median 1,469 words

Broker IntegrationTax, Accounting & Reporting

Get started

Clone it. Point
your agent at it.

The library needs no runtime and no service. It is Markdown and standalone Python, read by any tool that understands a SKILL.md. The repository ships instruction files for each of these:

  • Claude Code
  • GitHub Copilot
  • Codex CLI
  • Cursor
  • Gemini CLI
  • Aider
  • Cline
  • Windsurf
  • Continue
git clone https://github.com/HimanshuJ16/Algo-Trading-Skills.git
pip install -r requirements-dev.txt
python tools/validate_skills.py
python tools/run_all_tests.py