8x Productivity

An AI agent roster and harness for product engineering teams.

Eight specialized agents — each with distinct expertise, personality, and domain boundaries — that cover the full product development lifecycle: from requirements to architecture to implementation to QA to deployment.

---

The Team

Agent	Role	What They Do
Nova	CEO & Orchestrator	Researches the codebase, coordinates work, delegates to specialists
Kepler	Product Analyst	Translates business needs into testable requirements and acceptance criteria
Turing	Solution Architect	Designs technical solutions grounded in existing codebase patterns
Euclid	Idea Crystallizer & Spec Writer	Crystallizes fuzzy ideas via first-principles thinking; decomposes designs into executable implementation steps
Ada	Full-Stack Engineer	Writes production-quality code — frontend, backend, tests, infrastructure
Nebula	QA & Security	Reviews code for correctness, security vulnerabilities, and edge cases
Rosetta	Technical Writer	Produces accurate documentation that matches the codebase
Comet	SRE & DevOps	Verifies builds, manages deployments, and ships releases

---

How It Works

The agents follow an Ideation → Plan → Implement → Verify workflow:

Ideation (optional)     Plan                          Implement      Verify & Ship
───────────────────     ─────────────────────────     ───────────    ─────────────────
Euclid crystallizes     Nova researches               Ada builds     Nebula reviews
the idea                Kepler defines requirements                  Rosetta documents
                        Turing designs architecture                  Comet deploys
                        Euclid writes the plan

Nova detects when a task is fuzzy and routes to Euclid for ideation before research begins. Clear tasks skip straight to planning.

Each agent has hard boundaries — they do their job and nothing else. The architect does not write code. The engineer does not make product decisions. QA does not fix bugs. This separation prevents the role confusion that degrades output quality in single-agent workflows.

---

Installation

Install with Claude

If you have Claude Code installed, run this one-liner in your terminal. Claude will guide you through the setup interactively — asking which CLIs you use, whether to install globally or into a specific project, and handling everything from there.

claude "$(curl -fsSL https://raw.githubusercontent.com/polvera-org/8x-productivity/main/docs/AGENT_SETUP.md)"

Manual install

Clone the repo, install dependencies, and run the interactive setup wizard:

git clone https://github.com/polvera-org/8x-productivity.git
cd 8x-productivity
npm install
npm run setup

npm run setup is an interactive wizard that asks which CLIs you want to install for (Claude Code, OpenCode, Codex, OpenClaw, Agent Zero), whether to install globally or into a specific project directory, and handles all the copying.

---

Project Structure

src/
└── agents/              # Source of truth — gitagent-compatible format
    ├── ada/
    │   ├── agent.yaml   # Manifest: name, description, model, tools
    │   └── SOUL.md      # Agent prompt body
    ├── nova/            # (same structure for all 8 agents)
    └── ...

scripts/
└── build.ts             # Renders src/agents/ into CLI-specific output formats

.claude/agents/          # Generated — Claude Code format
.opencode/agents/        # Generated — OpenCode format
.openclaw/               # Generated — OpenClaw format
.a0proj/agents/          # Generated — Agent Zero format
AGENTS.md                # Codex CLI manifest

Agents are defined in the gitagent format: each agent lives in its own directory with an agent.yaml manifest and a SOUL.md prompt. The build script reads these and renders CLI-specific output for Claude Code, OpenCode, OpenClaw, Agent Zero, and Codex.

---

Agent Design

Each agent prompt is built on these principles:

• Single responsibility — One agent, one domain. No crossing boundaries.
• Distinct personality — Behavioral traits that make the agent effective in its role. Kepler's analytical rigor. Ada's craftsmanship. Nebula's adversarial mindset.
• Explicit contracts — Clear input/output formats so agents can hand off work cleanly.
• Hard boundaries — Every agent has a "What You Do NOT Do" section preventing scope creep.
• Tech-stack agnostic — Domain expertise without framework lock-in. They adapt to your codebase.

---

Adding or Editing Agents

1. Edit src/agents/<name>/SOUL.md for prompt changes, or src/agents/<name>/agent.yaml for metadata.
2. Run npm run build to regenerate all output formats, or npm run setup to rebuild and reinstall to your targets.