Manera

AI-powered SDLC automation pipeline. Implement, review, test, and fix code using Claude and Kimi agents.

What it does

Manera processes tasks from a JSON backlog through a full software development lifecycle:

Todo → Implement → Review → Test → Done
          ↓ FAIL      ↓ FAIL
        Fix → retry  Fix → re-test

Each stage is handled by a specialized AI agent:

• Implementer (Sonnet) — builds features from acceptance criteria
• Reviewer (Opus) — code review for quality, security, conventions
• Tester (Opus) — verifies acceptance criteria with unit, integration, and contract tests
• Fixer (Opus) — fixes issues found during review or testing

Tasks retry automatically (up to 5 attempts) before being marked as blocked.

Quick Start

npx manera

This launches an interactive CLI that guides you through setup and pipeline execution.

Prerequisites

• Node.js >= 18
• Claude Code CLI installed and authenticated

First Run

If no project.json exists in your current directory, Manera will walk you through creating one. This configures:

• Project name and tech stack
• Build and lint commands
• Documentation paths
• Testing configuration (optional dev server + browser tests)

Interactive CLI

  Manera  v2.0.0
  Project: My App  |  Backlog: tasks/backlog_tasks.json

  ? Main Menu
  > Run Pipeline
    View Status
    Reset Tasks
    Configure Models
    Pipeline Settings
    Switch Backlog
    Exit

Run Pipeline

• Continue — process the next Todo tasks sequentially
• Retry specific task — reset and re-run a single task
• Start from specific task — skip ahead to a particular task
• Doc-first phase — update project docs from an epic brief before coding

Configure Models

Override which model handles each agent role:

• Implementer, Reviewer, Tester, Fixer, Doc Updater
• Choose between claude-sonnet-4-5-20250929 and claude-opus-4-6
• Settings persist to .sdlc-rc.json

Pipeline Settings

• Toggle verbose output
• Switch CLI provider (Claude / Kimi)
• Set epic brief path for doc-first phase

Headless Mode

Run the pipeline without the interactive CLI:

npx tsx src/run-tasks.ts                           # Process from first Todo task
npx tsx src/run-tasks.ts --retry:TASK-001          # Reset and retry specific task
npx tsx src/run-tasks.ts --start-from:TASK-005     # Start from specific task
npx tsx src/run-tasks.ts --epic-brief:docs/epic.md # Run doc-first phase
npx tsx src/run-tasks.ts --cli-kimi                # Use Kimi for implementation

Project Configuration

All project-specific settings live in project.json:

{
  "project": {
    "name": "My App",
    "description": "A web application",
    "techStack": "Next.js, TypeScript, Tailwind CSS",
    "conventions": "Use functional components, prefer server components"
  },
  "commands": {
    "build": "npm run build",
    "lint": "npm run lint"
  },
  "docs": {
    "prd": "docs/prd.md",
    "solutionDesign": "docs/solution-design.md"
  }
}

Backlog Format

Tasks are defined in a JSON backlog file:

{
  "tasks": [
    {
      "id": "TASK-001",
      "name": "Add user login page",
      "status": "Todo",
      "description": "Create a login form with email and password fields",
      "acceptance_criteria": [
        { "criterion": "Login form renders with email and password fields", "met": false },
        { "criterion": "Form validates email format", "met": false }
      ]
    }
  ]
}

Testing Tiers

Task-Level (per task)

Unit → Integration → Contract

Story-Level (after all tasks in a story pass)

Regression → Smoke → Security → Performance → Accessibility → Exploratory → UAT

Browser-based tests (Smoke, Accessibility, Exploratory, UAT) use MCP Puppeteer when a dev server is configured.

License

MIT