{"slug": "shipwright-harness-open-source-autonomous-delivery-agent-for-claude-code-mit", "title": "Shipwright Harness – open-source autonomous delivery agent for Claude Code (MIT)", "summary": "App Vitals released Shipwright Harness, an open-source autonomous delivery agent for Claude Code under the MIT license. The tool enables developers to plan, build, review, and ship code tasks through a single pipeline, running on their own codebase with optional Kubernetes deployment.", "body_md": "**The open-source autonomous delivery agent for Claude Code.** A deployable cloud agent and the autonomous coding system that powers it — built on the Shipwright plugin, running on your own codebase.\n\n[\n](/app-vitals/shipwright/blob/main/assets/demo.gif)\n\n*Plan → build → review → ship — one task through the pipeline. (Illustrative.)*\n\nBrand vs. package:the project isShipwright Harness; the plugin/package you install is.`shipwright`\n\n```\n/plugin install shipwright@app-vitals/shipwright\n```\n\nRequires [Claude Code](https://www.anthropic.com/claude-code). Point it at your own repository — Shipwright is repo-agnostic.\n\n**Deploying the services to Kubernetes?** The `shipwright`\n\nHelm chart is published to a Helm repo on each chart version bump:\n\n```\nhelm repo add shipwright https://app-vitals.github.io/shipwright\nhelm install my-release shipwright/shipwright --namespace shipwright --create-namespace\n```\n\nSee [ docs/deploy-kubernetes.md](/app-vitals/shipwright/blob/main/docs/deploy-kubernetes.md) for end-to-end deployment guides (Minikube / GKE / EKS), and\n\n[for the published-repo flow and how publishing is triggered.](/app-vitals/shipwright/blob/main/docs/helm-repo.md)\n\n`docs/helm-repo.md`\n\nYou can run the **metrics dashboard locally today** — **offline by default**, with **no PostHog key, no accounts, and no database**. One copy-paste prompt sequences the two execution contexts (terminal shell + an in-session slash command) and opens the dashboard.\n\nPaste this into a **Claude Code** session:\n\n```\nSet up Shipwright Harness locally and open the metrics dashboard.\n\n1. In a terminal, run:\n     git clone https://github.com/app-vitals/shipwright.git && cd shipwright && ./scripts/quickstart.sh\n   This checks prerequisites, installs dependencies (task setup), and starts the\n   metrics dashboard in offline mode (no accounts or secrets needed). Leave it running.\n\n2. Inside this Claude Code session, install the plugin:\n     /plugin install shipwright@app-vitals/shipwright\n\n3. Open the dashboard in your browser:\n     http://localhost:3460/dashboard\n```\n\nStep 1 runs in your **terminal**; step 2 is a slash command that runs **inside the Claude Code session**. The dashboard comes up at [http://localhost:3460/dashboard](http://localhost:3460/dashboard).\n\nPrerequisites: [Claude Code](https://www.anthropic.com/claude-code), [git](https://git-scm.com/downloads), [Bun](https://bun.sh), and [go-task](https://taskfile.dev/installation/). Full details, the `QUICKSTART_SKIP_SERVE`\n\nCI guard, and the offline-default explanation live in [ docs/quickstart.md](/app-vitals/shipwright/blob/main/docs/quickstart.md).\n\n**Full dev stack**\n\nWant the complete local stack — metrics dashboard, admin UI, task-store, and the Shipwright agent running in Docker? This requires [tmux](https://github.com/tmux/tmux), [Docker](https://docs.docker.com/get-docker/), and a local [PostgreSQL](https://www.postgresql.org/) instance (port 5432). Paste this into a **Claude Code** session:\n\n```\nSet up the full Shipwright Harness dev stack locally.\n\nPrerequisites: tmux, Docker, PostgreSQL running on localhost:5432, Bun, go-task.\n\n1. In a terminal, clone and set up:\n     git clone https://github.com/app-vitals/shipwright.git && cd shipwright && task setup\n\n2. Copy the env example and add your auth token:\n     cp state/dev-agent.env.example state/dev-agent.env\n   Open state/dev-agent.env and set one of:\n     CLAUDE_CODE_OAUTH_TOKEN=<your token>   (run: claude /oauth-token)\n     ANTHROPIC_API_KEY=<your key>           (https://console.anthropic.com/ → API Keys)\n\n3. Launch the full stack (6-pane tmux session):\n     task stack\n\n   This opens a tmux session named \"shipwright\" with 6 panes:\n     metrics (:3460)  admin (:3001)  task-store (:3002)  agent (:3000)  chat  logs\n\n4. Open the dashboard in your browser:\n     http://localhost:3460/dashboard\n\n5. Inside a Claude Code session:\n     /plugin install shipwright@app-vitals/shipwright\n\nTo stop: tmux kill-session -t shipwright\n```\n\nTwo faces, one product:\n\n**The agent**— deploy it to your cloud (GitHub Actions or self-hosted). It does autonomous coding on your codebase, held to the** same review and test bar as human code**.** The system**— the autonomous coding system, built on the Claude Code: plan · build · review · metrics. Use it interactively inside Claude Code, or let the agent run it autonomously.`shipwright`\n\nplugin\n\nIt runs in **your** environment, on **your** codebase — you own it, it's MIT, and it's free.\n\nShipwright turns a feature idea into shipped, reviewed code through a sequence of Claude Code commands — each stage producing a durable artifact the next stage consumes:\n\n**Write a PRD** for your idea — a structured product spec ready for /plan-session.**Plan** the spec into a queue of well-scoped, dependency-ordered tasks (tracked as**GitHub Issues**— the queue lives where your team already works).** Execute**the next ready task — build, test, and open a PR.** Review**the PR with policy-controlled, inline feedback.** Ship**the merged change.\n\n**Free and open-source (MIT)**— the own-it alternative to closed, hosted coding agents. No rented infrastructure, no lock-in.** Runs in your environment, your cloud**— your code never leaves your control.** The same quality bar as human code**— tests land with the code, gated by a five-phase** test-readiness**pipeline, so an autonomous agent can be trusted.** Metrics on your own pipeline**— first-time-quality rate, estimation accuracy, and review-verdict trends, measured on your delivery.** Built on Claude Code**— we use it every day, and Shipwright extends it rather than replacing it.\n\n| Component | What it does | Status |\n|---|---|---|\nPlugin (the system) |\nThe `shipwright` toolchain you `/plugin install` — planning, queue-based execution, review, a test-readiness pipeline, and deploy commands. |\n✅ Available |\nMetrics dashboard |\nA stateless service that reads pipeline analytics (task throughput, CI first-pass rate, review verdicts, estimation accuracy) and renders a dashboard. Run locally with `task api` or `task ui` (offline mode, no secrets needed). |\n✅ Available |\nShipwright agent |\nA thin autonomous runner that drives the system on a schedule — pick the next ready task → build → ship a PR → forward metrics — deployable to GitHub Actions or self-hosted. | ✅ Available |\n\n```\n/shipwright:prd            → a product spec\n/shipwright:plan-session   → a dependency-ordered task queue\n/shipwright:dev-task       → build + test + open a PR for the next ready task\n/shipwright:review         → policy-controlled PR review\n/shipwright:patch          → address review findings / failing CI\n/shipwright:deploy         → merge + deploy\n```\n\nTasks are tracked as GitHub Issues, so the queue lives where your team already works.\n\n**See it live** — Shipwright builds itself. The [public metrics dashboard](https://proof.shipwrightharness.com/public/dashboard) shows live pipeline data for this repo: first-time quality, cycle time, estimation accuracy, and task throughput — all generated by the same agent and plugin you're reading about.\n\nShipwright Harness is live — plugin, metrics dashboard, and the Shipwright agent all ship with v0.1.0. See the [issues](https://github.com/app-vitals/shipwright/issues) for the live roadmap and upcoming improvements.\n\nThe metrics dashboard is runnable locally today — the [Quickstart](#quickstart) wraps this in one copy-paste prompt (`./scripts/quickstart.sh`\n\n). The underlying tasks:\n\n```\ntask setup      # bun install\ntask api        # start metrics dashboard in offline mode → http://localhost:3460/dashboard\ntask dev        # dev supervisor: starts metrics + Ctrl-C kills all children\ntask stack      # full dev stack in a tmux session (6 panes) — requires tmux\n```\n\n`task stack`\n\nbrings up a single tmux session (`shipwright`\n\n) with a 6-pane dashboard: **metrics** (SQLite, :3460), **admin** (CRUD API + UI, :3001), **task-store** (:3002), the **agent** with the dev `/chat`\n\nendpoint enabled (:3000), the **chat** REPL, and a scratch **logs** shell. It runs a Prisma `migrate deploy`\n\npreflight first so the admin service's Postgres schema is up to date; on macOS the preflight checks Postgres is reachable and offers to run the needed `brew`\n\n/`createdb`\n\ncommands if it isn't. `task stack`\n\nrequires `tmux`\n\n; if it isn't installed, use `task dev`\n\n(the no-tmux fallback that starts the metrics dashboard).\n\nSee [ docs/quickstart.md](/app-vitals/shipwright/blob/main/docs/quickstart.md) for the full onboarding prompt and offline-default behavior.\n\nShipwright Harness is a [Claude Code](https://www.anthropic.com/claude-code) plugin through and through — built on it, for it, and used with it daily. If you already run Claude Code, Shipwright is a `/plugin install`\n\naway.\n\nShipwright enforces a four-layer test architecture (unit / integration / smoke / e2e) across all three components. Layer boundaries, per-component run commands, speed budgets, and the test-isolation contract are defined in [ docs/test-readiness/test-system.md](/app-vitals/shipwright/blob/main/docs/test-readiness/test-system.md).\n\nAll configuration options — plugin env vars, `.shipwright.json`\n\nkeys, agent env vars, and policy fields — are documented in [ docs/configuration.md](/app-vitals/shipwright/blob/main/docs/configuration.md).\n\nIssues and discussion are welcome. See [ CONTRIBUTING.md](/app-vitals/shipwright/blob/main/CONTRIBUTING.md) for conventions and workflow, and our\n\n[. This repository is MIT-licensed and public — please keep contributions free of any proprietary or confidential material.](/app-vitals/shipwright/blob/main/CODE_OF_CONDUCT.md)\n\n`CODE_OF_CONDUCT.md`\n\n[MIT](/app-vitals/shipwright/blob/main/LICENSE) © 2026 App Vitals", "url": "https://wpnews.pro/news/shipwright-harness-open-source-autonomous-delivery-agent-for-claude-code-mit", "canonical_source": "https://github.com/app-vitals/shipwright", "published_at": "2026-06-30 01:59:31+00:00", "updated_at": "2026-06-30 02:22:27.601413+00:00", "lang": "en", "topics": ["ai-agents", "developer-tools"], "entities": ["App Vitals", "Shipwright Harness", "Claude Code", "Anthropic", "Kubernetes", "Docker", "PostgreSQL", "Helm"], "alternates": {"html": "https://wpnews.pro/news/shipwright-harness-open-source-autonomous-delivery-agent-for-claude-code-mit", "markdown": "https://wpnews.pro/news/shipwright-harness-open-source-autonomous-delivery-agent-for-claude-code-mit.md", "text": "https://wpnews.pro/news/shipwright-harness-open-source-autonomous-delivery-agent-for-claude-code-mit.txt", "jsonld": "https://wpnews.pro/news/shipwright-harness-open-source-autonomous-delivery-agent-for-claude-code-mit.jsonld"}}