Imagine each project in your portfolio having its own AI assistant that understands its unique architecture, coding conventions, and deployment patterns. The Per-Project AI Workflow Orchestrator is not a generic tool—it is a living configuration system that shapes itself to the DNA of every repository you touch.
Whether you work with Claude Code, Cursor, or Codex, this harness provides a 10-dimensional profiling engine that interviews your project, extracts its personality, and generates a bespoke AI coding environment. No two harnesses are ever the same.
Most AI coding tools treat every project identically, forcing square pegs into round holes. This harness flips the paradigm: the project dictates the tooling, not the other way around.
| Traditional Approach | This Harness |
|---|---|
| One config fits all | Per-project DNA extraction |
| Static rules | Adaptive profiling |
| Generic prompts | Context-aware instructions |
| Manual environment setup | Automated harness generation |
The result? Your AI assistant stops guessing and starts knowing—it understands your dependency trees, your testing philosophy, your deployment rituals, and even your team's communication patterns.
graph TD
A[Repository Clone] --> B[10-Dimension Profiler]
B --> C1[Architecture Detection]
B --> C2[Framework Analysis]
B --> C3[Convention Extraction]
B --> C4[Testing Philosophy]
B --> C5[Deployment Patterns]
B --> C6[Documentation Style]
B --> C7[Dependency Graph]
B --> C8[Performance Profile]
B --> C9[Security Posture]
B --> C10[Team Workflow]
C1 --> D[Harness Generator]
C2 --> D
C3 --> D
C4 --> D
C5 --> D
C6 --> D
C7 --> D
C8 --> D
C9 --> D
C10 --> D
D --> E[Claude Code Harness]
D --> F[Cursor Harness]
D --> G[Codex Harness]
E --> H[Custom AI Environment]
F --> H
G --> H
H --> I[Optimized Development Loop]
The profiler conducts a silent interview with your codebase, asking questions like: What frameworks do you trust? How do you handle errors? What does done look like? The answers form the blueprint for your custom harness.
Every project generates a unique profile. Here is what a generated configuration might look like for a Node.js microservices project with React frontend:
project_dna:
name: "microfrontend-platform"
detected_frameworks:
- "Next.js 14"
- "Express.js"
- "Jest + Playwright"
conventions:
error_handling: "centralized middleware"
code_style: "functional with type safety"
naming: "kebab-case for files, camelCase for functions"
testing_philosophy:
coverage_target: 85
integration_tests_required: true
mocking_preference: "nock for HTTP clients"
deployment_patterns:
strategy: "canary releases"
containerization: "Docker multi-stage builds"
ci_cd: "GitHub Actions with matrix builds"
team_workflow:
pr_size: "max 400 lines"
review_process: "two approvals required"
commit_convention: "conventional commits with scopes"This profile feeds into the harness generator, which produces distinct configurations for Claude Code, Cursor, and Codex—each optimized for that specific project context.
Running the profiler is straightforward. Navigate to any project and invoke:
npx harness-maker --profile --output ./ai-harnessDuring the profiling phase, you will see output like:
Starting 10-dimension interview with project "ecommerce-platform"...
✅ Architecture Detection: Microservices with event-driven communication
✅ Framework Analysis: Next.js 13 (App Router), NestJS, Prisma
✅ Convention Extraction: ESLint strict mode, Prettier 120-chars
✅ Testing Philosophy: TDD with Jest, Cypress for E2E
✅ Deployment Patterns: Kubernetes with Helm charts
✅ Documentation Style: Storybook + JSDoc
✅ Dependency Graph: 147 packages, 5 monorepo workspaces
✅ Performance Profile: Next.js ISR, Redis caching layer
✅ Security Posture: Helmet.js, rate limiting, input sanitization
✅ Team Workflow: Linear ticket references in commits
🎉 Profile Complete! Generating harness configurations...
✓ Claude Code harness: ./ai-harness/claude/
✓ Cursor harness: ./ai-harness/cursor/
✓ Codex harness: ./ai-harness/codex/
Your custom AI environment is ready. The harness knows your project better than you do.
Your harness runs anywhere your code lives. Compatibility matrix verified for 2026:
| Operating System | Claude Code | Cursor | Codex | Profiler |
|---|---|---|---|---|
| macOS Sonoma | ✅ Full | ✅ Full | ✅ Full | ✅ Full |
| macOS Sequoia | ✅ Full | ✅ Full | ✅ Full | ✅ Full |
| Windows 11 | ✅ Full | ✅ Full | ✅ Full | ✅ Full |
| Windows Server | ✅ Full | ✅ Full | ✅ Full | |
| Ubuntu 24.04 | ✅ Full | ✅ Full | ✅ Full | ✅ Full |
| Fedora 40 | ✅ Full | ✅ Full | ✅ Full | ✅ Full |
| Debian 12 | ✅ Full | ✅ Full | ✅ Full | ✅ Full |
| Arch Linux | ✅ Full | ✅ Full | ✅ Full | |
| Alpine (Docker) | ❌ Not Supported | ✅ Full | ✅ Full |
Full means all features are operational. Limited indicates the harness runs but some advanced profiling features may require additional dependencies.
Profiling Engine
- 10-Dimension Interview: Extracts architecture, framework, conventions, testing, deployment, documentation, dependencies, performance, security, and team workflow
- Silent Analysis Mode: Profiles without modifying any files or running destructive operations
- Incremental Updates: Detects changes since last profile and updates harness only where needed
- Multi-Language Support: Works with JavaScript, TypeScript, Python, Go, Rust, Java, Ruby, PHP, and C#
- Monorepo Awareness: Detects workspace structures and profiles each package independently
Harness Generation
- Three-Target Export: Simultaneously generates configurations for Claude Code, Cursor, and Codex
- Prompt Engineering: Creates context-aware system prompts that include project-specific conventions
- Rule Extraction: Converts linting rules, testing patterns, and coding standards into AI instructions
- Deployment Context: Includes CI/CD pipeline understanding in generated prompts
- Team Communication: Incorporates PR conventions, commit styles, and review processes
Runtime Features
- Responsive UI: Terminal-based interface adapts to window size and supports dark/light themes
- Multilingual Output: Generated prompts and instructions can be localized for international teams
- 24/7 Support Mode: Harness includes fallback behaviors for when AI models are unavailable
- Self-Healing Configs: Automatically fixes broken symlinks or missing dependencies
- Version Locking: Pin harness to specific AI model versions for reproducible behavior
Integration Capabilities
- OpenAI API: Direct integration with GPT-4o and o1 models for enhanced reasoning
- Claude API: First-class support for Claude 3.5 Sonnet and Claude 4 Opus (2026)
- Custom Endpoints: Plug in any OpenAI-compatible API endpoint
- Local Models: Support for Ollama and llama.cpp for air-gapped environments
- Git Hooks: Automatically install pre-commit and pre-push hooks with AI-enhanced checking
When you search for AI coding assistant profile generator or custom Claude Code configuration, this tool appears because it genuinely solves the problem those keywords describe. The harness is not just another wrapper—it is a project-specific AI environment optimizer that has been featured in discussions about context-aware coding tools and AI workflow orchestration.
Developers looking for per-project AI configuration, intelligent coding assistant setup, or Claude Code project awareness will find this harness delivers what generic tools cannot: true understanding of each codebase's unique characteristics.
The harness bridges the gap between your codebase and the most powerful AI models available:
// Example: Harness-generated OpenAI configuration
{
model: "gpt-4o-2026-01-01",
system_prompt: "You are an AI assistant for the 'payment-platform' repository. Follow these conventions: ...",
tools: [
{ type: "function", function: { name: "run_tests", parameters: { type: "object", properties: {} } } }
],
project_context: await harness.getContext()
}// Example: Harness-generated Claude configuration
{
model: "claude-3-5-sonnet-2026",
system: "You are working on a microservices architecture. The project uses: ...",
tools: [
{ type: "bash", name: "execute_command" },
{ type: "read", name: "read_file" },
{ type: "edit", name: "edit_file" }
],
max_tokens: 8192,
project_profile: await harness.getProfile()
}The harness automatically selects which model capabilities to emphasize based on your project's needs. For data-intensive projects, it biases toward models with larger context windows. For security-critical code, it enables additional safety guardrails.
The harness interface is built for adaptability. Whether you are working in a full-screen terminal on a 49-inch ultrawide monitor or a split-pane setup on a 13-inch laptop, the layout flows naturally. Tables collapse to lists, progress bars resize, and tooltips appear when space is constrained. Color schemes shift between light and dark modes based on your terminal preferences. Every interaction feels native to your environment.
Software engineering is a global discipline. The harness speaks your language—literally. Generated prompts and instructions support 12 languages out of the box:
| Language | Support Level |
|---|---|
| English | Full |
| Japanese | Full |
| Chinese (Simplified) | Full |
| German | Full |
| French | Full |
| Spanish | Full |
| Portuguese (Brazil) | Full |
| Korean | Full |
| Russian | Full |
| Arabic | Full |
| Hindi | Full |
| Dutch | Full |
Team documentation and code comments are used as input for language detection, ensuring the harness communicates with your AI in the same language your team uses.
This is not about human support agents—it is about architectural resilience. The harness includes:
- Graceful Degradation: If your primary AI model is unavailable, the harness automatically falls back to alternative models or local inference
- Redundant Configuration: Multiple provider configurations ensure continuous operation
- State Persistence: Work in progress is saved locally, so session interruptions do not lose context
- Auto-Recovery: Connection drops trigger automatic reconnection with context restoration
- Bandwidth Optimization: The harness intelligently caches common responses to reduce API calls during peak usage
Your development flow never stops, even when the cloud has a hiccup.
Important: This software is provided as-is without any express or implied warranties. The Per-Project AI Workflow Orchestrator generates configurations that interact with third-party AI services including but not limited to OpenAI API, Anthropic Claude API, and various code editors. Users are solely responsible for:
- Compliance with the terms of service of all third-party platforms
- Security implications of generated configurations, especially in production environments
- Cost management of API usage when using cloud-based AI models
- Data privacy and confidentiality when profiling proprietary codebases
- Reviewing and testing generated harnesses before deployment to critical workflows
The profiling engine analyzes your codebase to understand its structure and conventions. It does not transmit your code to any external service unless you explicitly configure it to do so. All profiling runs entirely on your local machine. The harness generation creates configuration files that may reference your project structure—review these files before sharing or committing them to version control.
The developers of this software assume no liability for any damages, data loss, or unintended consequences arising from the use of this tool. Always maintain backups of your original configuration files.
-
Install the package globally or use directly with
npx:npm install -g harness-maker
or
npx harness-maker
-
Navigate to any project:
cd your-project/ -
Run the profiler:
harness-maker --init
-
Your custom harness is ready at
./.harness/ -
Start your AI tool with the generated configuration.
This project is licensed under the MIT License. You are free to use, modify, and distribute this software for any purpose, commercial or personal, as long as the original copyright notice and permission notice are included in all copies or substantial portions of the software.
See the MIT License for full terms and conditions.
Copyright (c) 2026 Per-Project AI Workflow Orchestrator
Every codebase tells a story. This harness helps your AI listen.