For anyone who can define an options strategy and run Claude Code or Codex

A trading idea is not an engine.

Options OS gives your coding agent the definitions, specifications and libraries to build crypto-options engines, so it resolves the hard decisions with you instead of inventing them.

Buy Options OS · $149

See how it works

$149 one-time · no subscription · private GitHub invitation after payment

intent.md

Example · 7 accepted·3 open

01
01
market idea
market idea
accepted
accepted
02
02
data quality
data quality
accepted
accepted
03
03
signals
signals
accepted
accepted
04
04
instruments
instruments
accepted
accepted
05
05
sizing
sizing
accepted
accepted
06
06
orders
orders
open
open
07
07
hedging
hedging
open
open
08
08
settlement
settlement
accepted
accepted
09
09
failures
failures
open
open
10
10
operator controls
operator controls
accepted
accepted

Translation starts only when every readiness gate passes. Code starts after the specifications are accepted.How the intent works →

How an engine gets built

Staged, accepted, verified: before and after code

Each stage closes with an accepted artifact, and the next stage starts from it.

Options OS agent build progressing through Define, Translate, Architect, Contract and plan, Execute, and Verify

Define—Complete the trading engine intent with a web-based reasoning model. Unresolved decisions stay marked, not filled in.
Translate/translate-domain-intentTurn the intent into engine, operations and scenario specifications. Gaps go back to the intent; the agent does not invent policy.
Architect/architect-systemSet component boundaries, who owns state, where side effects happen and which libraries and venue binding the engine uses.
Contract and plan/write-buildable-specification · /create-implementation-planWrite implementation-ready specifications, then plan dependency-ordered packets the agent builds one at a time.
Execute/execute-implementation-planBuild the packets in order; each closes with tests, invariant checks, conformance and recorded commands and outcomes.
Verifypre-commit run --all-filesRun Ruff, Mypy, package tests, architecture and invariant enforcement, and control tests.

The problem

Every missing decision gives the agent room to invent financial meaning

A strategy sentence leaves decisions open. A coding agent cannot build or operate the engine without answering them, so it guesses.

01 · market ideaWhat exactly counts as “high volatility”?

02 · data qualityWhich feeds, and what happens when they go stale?

03 · signalsWhat triggers entry, and what cancels it?

04 · instrumentsWhich expiry and which strikes?

05 · sizingHow large, against which risk limit?

06 · ordersWhat happens after a partial fill?

07 · hedgingWhen does it re-hedge delta, and with what?

08 · settlementHow are costs, PnL and settlement reconciled?

09 · failuresWhat happens on disconnect or restart?

10 · operator controlsWhat can you see, pause and stop?

Without Options OS	With Options OS
The agent invents the financial decisions a strategy leaves open	The intent surfaces every decision before the build
Rebuild options infrastructure around every strategy	Compose an existing options-native foundation
Agents infer boundaries from scattered code	Specifications and contracts define the system
Venue behavior leaks into engine logic	Engine code consumes normalized venue facts, as the implemented Deribit adapter shows
Generated code is accepted on appearance	Verification is part of the build path

WithoutThe agent invents the financial decisions a strategy leaves open

WithThe intent surfaces every decision before the build

WithoutRebuild options infrastructure around every strategy

WithCompose an existing options-native foundation

WithoutAgents infer boundaries from scattered code

WithSpecifications and contracts define the system

WithoutVenue behavior leaks into engine logic

WithEngine code consumes normalized venue facts, as the implemented Deribit adapter shows

WithoutGenerated code is accepted on appearance

WithVerification is part of the build path

What you get

A foundation the agent composes, not rebuilds

Financial meaning is defined once across units, instruments, Greeks, costs, risk, orders and settlement, and every engine uses the same definitions.

Reusable libraries21

Matching specifications21

Implemented adapterDeribit

Agent skills24

End-to-end workflows5

Coding agents2

Intent and templates

Trading engine intent plus engine, operations and scenario specification templates. Define the strategy, risk limits, entry and exit rules, and expected behaviour before the agent starts building.

Options-native foundations

21 Python libraries covering trading models, options mathematics, portfolios, decisions, infrastructure and exchange integration, each with its own specification and tests.

Agent controls

Skills, workflows, hooks and tools for Claude Code and Codex. Repository checks stop the agent weakening tests or claiming verification it did not run.

Architecture, libraries and specifications in the docs →

Fit

Options OS supplies the foundation. You define the engine

Owner	Area
Options OS	Options-native semantics, libraries and specifications
Options OS	Implemented Deribit exchange adapter
Options OS	Trading engine intent and specification templates
Options OS	Claude Code and Codex skills, workflows, hooks and tools
Options OS	Invariants, tests and architecture enforcement
You	Strategy, trading signals and PnL mechanism
You	Risk limits, lifecycle and operating policy
You	Hosting, deployment, credentials and capital
You	Operational validation and live approval
You	Review of specifications and generated code; Python experience helps