The Stella Protocol

PM-first AI development protocol. One mind, many satellites.

Stella is a structured methodology for Product Managers who build 100% with AI. Instead of calling independent AI agents, you are Stella — the original mind — and AI operates as specialized extensions of your thinking called Satellites.

Why Stella?

Most AI development frameworks are built for developers. Stella is built for the PM who owns the product vision and uses AI to execute everything else — architecture, code, tests, deployment, documentation.

• No commands to memorize. Satellites activate from natural conversation, not slash commands.
• Built-in governance. Scope enforcement and veto protocols are always on — not optional tools you remember to invoke.
• Persistent project brain. A structured brain/ directory tracks decisions, state, and scope — committed to git, reviewable in PRs.

Quick Start

npx stella-protocol install

This will:

1. Install 5 phase-based Agent Skills into your AI tool
2. Initialize a brain/ directory (Punk Records) in your project
3. You're ready — start by telling your AI what you want to build

The Protocol

Phases

PHASE 0: IDEATE    →  Think about the problem space
PHASE 1: DEFINE    →  Requirements, architecture, UX, PRD
PHASE 2: BUILD     →  Implementation, deployment
PHASE 2.5: ITERATE →  Post-launch feedback loop
PHASE 3: CLOSE     →  Documentation, launch, go-to-market

Two Tracks

Satellites

Satellites are specialized facets of your extended cognition. They activate from conversational context:

Governance (Always On)

Buster Call — Any satellite can halt progress when it identifies critical issues. Severity: CONCERN → WARNING → BUSTER CALL. At BUSTER CALL level, work stops until you resolve it.

Cipher Pol — Continuous scope monitoring against the approved PRD. Severity: INTEL → ALERT → INTERCEPT. Drift is detected and flagged automatically. You decide whether to approve or reject.

Punk Records (Project Brain)

Every project gets a brain/ directory — git-committed, PR-reviewable:

brain/
├── log-pose.md        # Where are we? Phase, track, blockers, active work
├── architecture.md    # Tech decisions with rationale
├── vivre-cards.md     # Append-only decision log
├── ideas.md           # Unprocessed idea backlog
└── prd-*.md           # PRD documents (created as needed)

Agent Skills

Stella packages satellites into 5 phase-based skills:

Compatible with Claude Code, Cursor, and any tool supporting the Agent Skills open standard.

Philosophy

You are not "talking to an AI agent." You are thinking, and specialized facets of your extended cognition activate to help you execute.

Stella inverts the typical framework model. You don't call agents — you think out loud, and the right expertise surfaces. The governance protocols aren't tools you invoke — they're constitutional rules that every satellite enforces.

This is a protocol, not a product. The methodology in protocol/ is human-readable markdown. You can apply it manually without any tooling. The skills/ layer packages it for AI tools, but the protocol stands on its own.

Project Structure

stella-protocol/
├── protocol/          # The methodology (human-readable markdown)
│   ├── stella.md      # Master protocol
│   ├── satellites/    # Satellite definitions
│   ├── governance/    # Buster Call + Cipher Pol
│   └── tracks/        # Grand Line + East Blue
├── punk-records/      # Brain directory templates
├── skills/            # Agent Skills (AI-consumable)
├── cli/               # CLI installer
└── docs/              # Documentation

Example

See the full walkthrough — a worked example building "LinkShelf" (a bookmark manager) through the entire Stella cycle, with populated brain/ files showing what each phase produces.

License

MIT

Author

Aditya Uttama — Product Manager who builds 100% with AI.