Ship Autonomous MVPs with Claude Code

Shiploop's /plugin installs an autonomous delivery loop for Claude Code, running parallel DEV agents to deliver an MVP with contracts and evidence gates.

AAAla SMITH & AI Research Desk·1d ago·3 min read··5 views·AI-Generated·Report error

Source: github.comvia hn_claude_code, gn_claude_community, devto_claudecode, reddit_claudecodeMulti-Source

How do I install and use Shiploop to run autonomous parallel Claude Code agents for MVP delivery?

Shiploop is an autonomous delivery loop for Claude Code that turns an idea pitch into an MVP. It orchestrates parallel DEV agents, uses contracts for escalation, and gates on evidence (tests pass). Install via `/plugin marketplace add BechsteinDigital/claude-shiploop`.

TL;DR

Shiploop runs parallel Claude Code agents from a single pitch, autonomously delivering an MVP with contracts, gates, and evidence rules.

What Changed — Autonomous Delivery Loop for Claude Code

Claude Code /loop — How I Create New Native Autonomous Loops That Work ...

Shiploop is a new open-source skill suite for Claude Code that turns a single idea pitch into a fully autonomous MVP delivery loop. Instead of manually steering Claude Code through every step, you pitch your idea once, answer a brief onboarding interview, and the suite handles everything from research to parallel development to a gated MVP.

The project is available at github.com/BechsteinDigital/claude-shiploop and installs as a Claude Code plugin via /plugin marketplace add BechsteinDigital/claude-shiploop.

What It Means For You — From Toolbox to Operating Model

Most Claude Code skill collections are toolboxes — they give you tools, but you still drive. Shiploop is an operating model. It introduces:

Autonomy with a contract: Escalation to you only via explicitly agreed criteria. Everything else is decided and logged.
Focus as mechanics, not appeals: Value filter, cooling-off, idea-chain rule, extension budget — prohibitions alone don't keep an autonomous loop on course.
Claims need evidence: Documentation may never claim more than code + tests prove.
Parallelism via claim zones: Disjoint file zones per package; worktree isolation when in doubt.

This is a significant shift for teams who want Claude Code to operate more like a junior dev team than a coding assistant. The demo replays a real run: pitch → interview → setup → 3 loop cycles (2 DEV agents in parallel) → MVP gate (standup CLI, 36/36 tests).

Try It Now — Install and Run Your First Autonomous Loop

Installation (Recommended: Plugin Method)

Inside Claude Code, run:

/plugin marketplace add BechsteinDigital/claude-shiploop
/plugin install claude-shiploop@bechstein-digital

Or via install script:

git clone https://github.com/BechsteinDigital/claude-shiploop.git
cd claude-shiploop
./install.sh /path/to/your/project

Running Your First Loop

Start a Claude Code session inside your target project.
Pitch your idea — project-onboarding picks it up from there.
Answer the onboarding interview once (research, stack, scaffold, backlog).
The suite runs autonomously: it researches, picks a stack, scaffolds the project, cuts work packages, and orchestrates role agents in parallel until the MVP gate.

Project Structure (What Gets Created)

Shiploop creates a structured project with contracts and evidence rules:

project/
  BRIEF.md          # product brief + core contract + autonomy contract
  PROFILE.md        # stack, verified commands, quality rules (single source for commands)
  STATE.md          # single state store (WIP, milestone, budget, cycle)
  DECISIONS.md      # ADR-light log of autonomous decisions
  IDEAS.md          # idea funnel with triage rules
  LEARNINGS.md      # retro distillate of this project (gate requirement)
  backlog/          # WORK cards (incl. claim zone + complexity)
  log/              # research, audit, and cycle logs

Key Principles to Understand

One source per truth: Commands only in PROFILE.md, state only in STATE.md — no snapshot duplicates that drift.
Parallelism: Agents work on disjoint claim zones to avoid conflicts. Use worktree isolation when in doubt.
Escalation gates: The suite only escalates to you on criteria you explicitly agreed to in the onboarding contract.

Source: github.com

Source: gentic.news · 1d ago · author=Ala SMITH · citation.json

AI-assisted reporting. Generated by gentic.news from multiple verified sources, fact-checked against the Living Graph of 4,300+ entities. Edited by Ala SMITH.

Following this story?

Get a weekly digest with AI predictions, trends, and analysis — free.

AI Analysis

Shiploop changes how you should think about using Claude Code for greenfield projects. Instead of iterating manually with prompts, you define the contract upfront and let the system run. This is particularly valuable for: 1. **Rapid prototyping**: Pitch an idea, get an MVP you can review. The parallel agents cut delivery time significantly. 2. **Structured autonomy**: The contract-based escalation means you're not babysitting every step. You only get pulled in when the system hits a decision boundary you defined. 3. **Evidence-based delivery**: Documentation can't claim more than tests prove. This reduces the risk of "works on my machine" syndrome in autonomous development. To get started immediately: install the plugin, create a new project directory, and pitch a small idea (e.g., "a CLI tool that tracks daily standup notes"). Run through the onboarding interview and let the loop run. Review the STATE.md and DECISIONS.md to see how the system operates. This is the fastest way to understand the model and decide if it fits your workflow.

#open source #claude code #plugin #autonomous agents #mvp

This story is part of

Claude Code's Campus Conquest Flips Anthropic's Talent Pipeline, Leaving Google's Academic Edge in Doubt

Viral adoption at MIT and Stanford transforms Claude Code from product into recruiting funnel, threatening Google's long-held research talent dominance

Compare side-by-side

Claude Code vs Shiploop

→

Mentioned in this article

Shiploop Claude Code BechsteinDigital

Enjoyed this article?