The governed autonomy standard · {{ scoreBadge }}
Modonome

The Governed Autonomous
Engineering Loop.

Modonome makes steady, provable progress on technical debt a normal part of engineering. It opens small, test-backed pull requests and keeps the maker, the checker, and the merge authority as separate roles, so every change is reviewed and proven before it lands.

github.com/nateshpp/modonome
Run the simulator → ★ Star on GitHub
MIT licensed·Runs inside your repo·Private by design·You decide when it acts
Why it matters

Built for the real balances of autonomous engineering.

Autonomy is only useful when it is trustworthy. Modonome is built to hold the tensions that matter in steady balance.

{{ bl.left }} {{ bl.right }}
{{ bl.title }}
{{ bl.body }}
How it works · the eight-step loop

One small change, from queue to merge.

Each card shows who acts and when. Modonome plays the maker and checker; your CI runs the gates; you stay the owner and merge authority. The tag on each step names the actor.

Modonome is a prompt and a set of scripts. It runs inside a harness you already have, whether that is a coding agent, a CI job, or a human session that loads the prompt.

{{ sp.num }} {{ sp.actor }}
{{ sp.title }}
{{ sp.body }}
Capabilities

Capabilities that read clearly to product teams.

Each one is small enough to trust and specific enough to verify. Together they move the work that usually waits.

{{ ft.iconEl }}
{{ ft.title }}
{{ ft.body }}
Structural by design

The controls are real code in your pipeline.

A clever prompt can talk another prompt into almost anything. A CI gate that runs outside the agent’s write scope answers only to your pipeline. The ratchet, the validators, and the drift guard all live there, so the guarantees hold steady even when a prompt tries to bend them.

Autonomy turns on only when an owner sets MODONOME_ARMED. Until then it stays at rest.
{{ g.icon }}
{{ g.title }}
{{ g.body }}
Governed-autonomy simulator

Pick a change. Watch the loop decide.

Send one change down the pipeline and watch where it lands and which control holds the line. A held or escalated change loops back for rework, and every outcome matches a real AgentProof scenario.

Choose a change to send
{{ grp.label }}
CHANGE {{ activeLabel }}
{{ nd.glyph }}
{{ nd.title }}
{{ nd.sub }}
{{ nd.scope }}
↺ {{ loopCaption }}
{{ verdictTag }} {{ verdictTitle }}

{{ verdictDetail }}

AgentProof benchmark

Sixteen adversarial scenarios. Machine-verified, every one.

AgentProof is a standalone, language-agnostic benchmark. Each of the sixteen scenarios runs a real adversarial attempt against the controls and confirms it is caught, so a passing score is verifiable evidence rather than a claim. Any agentic system can run it, which makes it a practical conformance test for the wider field, and beyond Modonome.

Conformance: HARDENED = 16/16 gate-integrity scenarios · PARTIAL = 12–15 · UNHARDENED ≤ 11
{{ apScore }}
{{ apLevel }}
{{ p.id }}
{{ p.t }}
$ node agentproof/runner.mjs → Score: {{ apScore }} · Level: {{ apLevel }} {{ syncNote }}
Living-system simulator

Watch your code and Modonome improve together.

Trigger an event and follow it across the boundary. Some changes land in your code, some sharpen Modonome’s rules, and some upgrade Modonome itself, all inside the same repository.

Lands in your code Sharpens the rules Upgrades Modonome | The boundary is the line the agent can’t cross, enforced in CI
{{ evoConnectors }}
Your application
Any stack · your existing process
{{ rn.icon }}
{{ rn.title }}
{{ rn.sub }}
{{ rn.readout }}
BOUNDARY
Modonome · in the same repo
Runs inside your repository · {{ engineVersion }} · {{ armedLabel }}
{{ en.icon }}
{{ en.title }}
{{ en.sub }}
{{ en.readout }}
Trigger {{ evoStepLabel }}
{{ evoTitle }}
{{ evoSummary }}
{{ evoOutcome }}
Activity
{{ lg.where }} {{ lg.text }}
The bigger picture · roadmap direction

Many repositories. One governed knowledge network.

Here is where Modonome is headed. A pattern proven in one repository can travel to another as verifiable evidence, and the receiving repository adopts it only after it passes that repository’s own checks. Every repository keeps its own controls, so trust is earned locally at each step rather than granted by a central authority.

{{ networkConnectors }}
Governed mesh
Application
Knowledge Network
{{ netShared }} patterns shared
Node · TS
Payments API
service · 41k LOC
◆ Modonome · armed
Python
Risk Models
ml pipeline · 28k LOC
◆ Modonome · armed
Java
Core Banking
platform · 210k LOC
◆ Modonome · armed
.NET
Claims Portal
web app · 63k LOC
◆ Modonome · armed
Go
Edge Gateway
service · 19k LOC
◆ Modonome · armed
COBOL
Ledger Core
mainframe · 1.2M LOC
◆ Modonome · armed
1
A pattern is earned locally
One repo proves a fix or safeguard against its own tests and gates.
2
Shared across the boundary
It travels the governed mesh as verifiable evidence, never as raw write access.
3
Adopted only if it re-passes
Each repo re-verifies the pattern under its own rules before it ever lands.
Where it is going

An open roadmap, shaped in public.

v0.2 and v0.3 are planned and tracked in the repository. The milestones beyond them describe the direction we want to explore together, and the community is welcome to help shape what comes next.

{{ ms.phase }} {{ ms.status }}
{{ ms.title }}
{{ ms.body }}
Get started

Start in a minute. Stay in control the whole way.

Three short steps take you from a safe preview to real, verifiable evidence. Nothing changes in your repository until you choose, and autonomy waits for you to switch it on.

{{ q.n }}
{{ q.title }}
{{ q.t }}
modonome · zsh
# Try it safely, this changes nothing
$ npx modonome dry-run .
✓ detected: node · typescript · jest · github-actions
→ 5 small changes proposed · 0 files written
# Scaffold local state (still disabled)
$ npx modonome scaffold . --write
✓ .modonome/ created · autonomy_enabled: false
# Prove the guarantees yourself
$ node agentproof/runner.mjs
Score: {{ apScore }} Level: {{ apLevel }}
Build the cause together

Secure, autonomous engineering for the assets that matter most.

An open standard can help enterprises trust autonomous engineering and steadily pay down the debt in their most critical applications. We intend to propose AgentProof to the OWASP Agentic Working Group, the OpenSSF Securing Software Repositories WG, and the AAIF, and the work is open to everyone.

Adopt and share
Run it on a repository, then tell the community what worked and what to improve.
Add a scenario
Bring a new adversarial case to AgentProof and help raise the bar for everyone.
Shape the standard
Join the discussion on the spec and the roadmap, and help guide where it goes next.