Auterix

One workflow across 21 native AI tools. Universal agent standard.

Auterix is a deterministic, tool-neutral autonomous context protocol and multi-agent engineering workflow. Projects define shared architectural invariants, task lifecycles, and verification gates once, while Auterix compiles, verifies, and locks them natively across 21 modern AI coding assistants and platforms — including Cursor, Claude Code, Google Antigravity, Windsurf, GitHub Copilot, Devin, Aider, OpenHands, Cline, Roo Code, CodeRabbit, Zed, Trae, and more.

⭐ If Auterix saves you engineering hours or prevents AI regressions, please star the repository — it helps more engineers discover the project!
📖 Explore the Documentation Hub (docs/) | 60-Second Quickstart | Free vs. Pro vs. Agency Guide
⚡ Live Web Studio: Configure, auto-detect your stack, test hallucination diffs, and export unified AI configs visually at auterix.vercel.app.
📦 Commercial Pro Suite: Production stacks, runnable templates, pre-commit AI guardrails, turnkey PR review bots, and 6 specialized role personas available at Gumroad ($14).

⚡ What Makes Auterix Different?

| Challenge with Raw Prompts / Rules | How Auterix Solves It | | :--- | :--- | | Tool Fragmentation: Maintaining separate .cursorrules, CLAUDE.md, and .agents/ leads to out-of-sync rules. | Single Canonical Source: Auterix synchronizes all 21 client adapters with zero drift. | | Silent Regressions: AI claims "task done" without running tests or verifying constraints. | Deterministic Verification: Staged writer guards, content-addressed hash locks, and pre-commit checks. | | Context Window Rot: Massive prompt files cause LLM attention degradation and blow token budgets. | Progressive Disclosure: Hierarchical 3-tiered context loading (Core Invariants ➔ Stack Blueprints ➔ Task Contracts). | | Lost Chat State: Switching between Cursor (inline) and Claude/Antigravity (deep refactoring) resets progress. | Cross-Agent Task Contracts & Shared Memory: Formal state machine & .ai/memory.md persist institutional decisions across models. | | Architectural Drift: Team members merge unconstrained AI code that violates database RLS or leaks secrets. | Turnkey CI Compliance: Headless GitHub Actions & PR bots audit rules and block non-compliant diffs in <1.5s. |

🩺 AI Readiness Scorecard (npx auterix doctor)

Scan any repository and receive a scored diagnostic report (0–100%) checking 8 core AI safety dimensions with actionable fix commands:

┌──────────────────────────────────────────────────┐
│  🏥  AUTERIX DOCTOR — AI Readiness Report        │
├──────────────────────────────────────────────────┤
│  ✅ Adapter Sync ..................... 12.5 / 12.5│
│  ✅ Schema Distillation .............. 12.5 / 12.5│
│  ✅ Memory File ...................... 12.5 / 12.5│
│  ✅ Pre-Commit Hook .................. 12.5 / 12.5│
│  ✅ Protected Boundaries ............. 12.5 / 12.5│
│  ✅ Task Tracking .................... 12.5 / 12.5│
│  ✅ Secret Leak Scan ................. 12.5 / 12.5│
│  ✅ Roles Defined .................... 12.5 / 12.5│
├──────────────────────────────────────────────────┤
│  SCORE: 100 / 100  EXCELLENT                     │
└──────────────────────────────────────────────────┘

Run in any project:

npx auterix doctor

Or export as structured JSON for CI pipelines:

npx auterix doctor --out report.json

🎭 Modular Role Personas (.ai/roles/)

Instead of bloated monolithic prompts that confuse LLMs, Auterix generates focused, modular role personas:

• architect.md: System topology, high-level design, database schema invariants, boundary enforcement.
• engineer.md: Idiomatic, typed implementation adhering strictly to project architectural rules.
• reviewer.md: Rigorous PR auditing, vulnerability detection, and invariant compliance checks.
• verifier.md: End-to-end acceptance testing, regression suites, and proof-of-correctness.

(Pro Suite unlocks 6 additional specialized roles: Security Auditor, Database Architect, QA Engineer, API Designer, Performance Engineer, and DevOps Engineer).

🧠 Cross-Agent Shared Memory & Git Hooks

Keep institutional memory intact across Cursor, Claude, Antigravity, and Windsurf so discarded libraries and bad patterns are never re-introduced.

1. Automatic Post-Commit Ingestion

Simply add --ai-note <category> to any git commit message:

git commit -m "Migrate from Redux to Zustand --ai-note architecture"
git commit -m "Sanitize user inputs before DB queries --ai-note security"

The installed Git post-commit hook automatically appends the decision, category, author, and timestamp into .ai/memory.md.

2. Manual CLI Logging

npx auterix memory --category architecture --note "We use Drizzle ORM + pgvector for RAG"
npx auterix memory --category anti-patterns --note "Do not use client-side Supabase service role keys"

🎨 Single-Source Formatter Config (.formatter/)

AI assistants drift on formatting as much as on architecture — one commit in tabs, the next in spaces. npx auterix init scaffolds .formatter/FORMATTER_CONFIG.md alongside your project (edit once, in plain KEY=VALUE form) plus a generator that keeps .prettierrc.cjs and .prettierignore in sync with it:

node .formatter/sync.cjs   # regenerates .prettierrc.cjs + .prettierignore

The generated .prettierignore is a blacklist, not a whitelist — it stays correct in monorepos with multiple source roots (apps/*/src, packages/*/src), where a whitelist-style ignore file (/* + !/src/) silently stops formatting everything outside a single root-level src/. Consistent with Auterix's portable core selecting nothing on your behalf: this file is yours from the first init, never overwritten by later syncs, and installing it doesn't touch your package.json.

🛡️ Rich CI Compliance Reports (action.yml)

Add turnkey AI rule compliance audits to your GitHub PRs at $0.00 compute cost. Auterix formats rich Markdown tables directly into $GITHUB_STEP_SUMMARY:

- name: Auterix AI Workflow Audit
  uses: SapanMozammel/auterix@main

Pull Request Summary Output:

| Check | Status | Details | | :--- | :--- | :--- | | SHA-256 Integrity | ✅ Pass | 21 managed files verified | | Manifest Valid | ✅ Pass | Schema v1 compliant | | Protected Boundaries | ✅ Pass | No unauthorized mutations | | Context Validation | ✅ Pass | All targets accessible |

🗄️ Database Schema Extractor (npx auterix extract-schema)

Extract token-compact Markdown schemas from Drizzle ORM, Prisma, or SQL DDL for zero-hallucination query generation:

npx auterix extract-schema --file blueprints/database/drizzle.schema.ts

Reduces prompt token overhead by >70% compared to dumping raw schema files into context.

Supported AI Tools & Platforms (21 Total)

Auterix provides native discovery and rules for:

• Native AI IDEs: Cursor (.cursor/), Windsurf (.windsurfrules), Trae (.trae/), Zed (.zed/)
• Autonomous Terminal Agents: Claude Code (CLAUDE.md), Google Antigravity (.agents/), Aider (CONVENTIONS.md), OpenHands (.openhands_instructions), Devin (DEVIATION.md), OpenAI Codex (AGENTS.md)
• Agent Extensions & Modes: GitHub Copilot (.github/copilot-instructions.md), Cline (.clinerules), Roo Code / Kilo Code (.roomodes), Continue.dev (.continue/), Augment Code, Tabnine
• AI PR Review & CI Bots: CodeRabbit (.coderabbit.yaml)
• Generative App Builders: v0 by Vercel, Bolt.new, Lovable.dev, Replit Agent

Read the 21 AI Tools Playbook for full setup runbooks.

Quickstart

Option A: Interactive CLI Wizard (Fastest)

Run directly inside any repository:

npx auterix@latest

Launches an interactive ANSI wizard to pick your tools, select your production stack, and install pre-commit guardrails and post-commit memory hooks.

Verify rule integrity at any time:

npx auterix check

Run health diagnostics:

npx auterix doctor

Option B: Visual Web Studio (In-Browser)

Open the Auterix Web Studio to test interactive hallucination diffs, auto-detect your stack, and export unified multi-tool bundles.

Option C: VS Code & Cursor Extension

Install the official Auterix extension from extensions/vscode/ for live status bar telemetry ($(shield) Auterix: 21 Tools Synced), real-time drift alerts, and 1-click adapter re-sync.

Production Stack Profiles (Pro Blueprints)

Auterix includes battle-tested production profiles available interactively in the Web Studio and packaged in the Auterix Pro Suite:

• Next.js 15 App Router + Supabase RLS: Strict Server Component boundaries, Zod-validated Server Actions, and mandatory PostgreSQL Row Level Security.
• FastAPI + Async SQLAlchemy 2.0: Pure async connection hygiene, Pydantic v2 schema isolation, and layered repository patterns.
• Enterprise Node.js 22 + Clean Architecture: Typed domain error boundaries, request correlation IDs, and transaction isolation.
• React Native & Expo 51+ Mobile: Platform-defensive guards, safe area wrappers, and atomic Zustand selectors.
• AI & Autonomous Agent Pipelines: Structured JSON outputs, evaluation harnesses, and exponential retry fallbacks.

🧪 Comprehensive Test Suite (Zero Dependencies)

node --test

Auterix includes 91 automated unit tests (0 external npm dependencies, running on Node.js >=22 native test runner):

• Deterministic, content-addressed SHA-256 bundle digests.
• Diagnostic 8-check health scoring (lib/doctor.mjs).
• Automated commit message memory parsing (lib/memory.mjs).
• Atomic staging, rollback guards, and preserved concurrent edits.
• Task state machine contracts, cycle detection, and deferral expiration.
• Symlink traversal and secret-file leak protections.

Review provenance, architecture, and context contract.

License & Attribution

Created by Sapan Mozammel.

Original maintained code and guidance are released under the MIT License.