Drydock

Drydock is an open-source plugin for Claude Code that turns it into a full product team. You describe what you want in plain English, and a single orchestrator routes the work to 19 specialized agents — covering product, architecture, UX, engineering, QA, security, DevOps, SRE, compliance, docs, and go-to-market. The agents research, build, test, secure, document, and launch a real system. You stay in the strategist's seat and approve at three checkpoints; the agents do the work in between.

Each agent is invocable as a drydock:<skill> skill. Throughout these docs, "agent" and "skill" mean the same thing: a specialized worker the orchestrator routes to.

:::info Version This documents Drydock 2.5.0. Source: github.com/sundarshahi/drydock (MIT licensed). :::

What Drydock is

Drydock is a team of 19 agents coordinated by one orchestrator:

• 15 agents run as isolated subagents (agents/*.md), each in its own context.
• The orchestrator (drydock) plus 3 planning agents — Product Manager, Solution Architect, and Polymath — run in-context as skills.

You don't manage the agents directly. You describe a goal, pick an autonomy level (how often Drydock checks in), and the orchestrator infers what work is needed and routes it. It installs entirely inside Claude Code — no local clone required.

The idea-to-launch pipeline

A full build flows through six phases. The agents work autonomously inside each phase; you weigh in only at the three approval gates.

Your idea
   │
   ▼
DEFINE   → requirements (BRD) · architecture + API · UX design system & flows
   │   ◆ Gate 1: approve requirements
   │   ◆ Gate 2: approve architecture
   ▼
BUILD    → backend · frontend · DevOps · QA · security · code review · SRE   (parallel)
   │
   ▼
HARDEN   → tests · security audit · code review · container build            (parallel)
   │
   ▼
SHIP     → IaC + CI/CD · SRE · remediation
   │   ◆ Gate 3: approve production readiness
   ▼
LAUNCH   → marketing/growth · sales · customer success                       (parallel)
   │
   ▼
SUSTAIN  → docs · project-specific skills · compound learning
   │
   ▼
A tested, secured, documented, deployed system — ready to launch.

The three human approval gates are non-negotiable checkpoints:

Gate	You approve
Gate 1	Requirements (the BRD)
Gate 2	Architecture
Gate 3	Production readiness

:::note Not every request runs the whole pipeline Drydock picks an execution mode to match what you asked for. There are 14 modes — Full Build runs all 19 agents through every phase; the other 13 (Feature, Harden, Pentest, Compliance, Ship, Test, Review, Architect, Design, Document, Explore, Optimize, Launch) route to a subset with fewer or zero gates. A Custom fallback shows a menu when nothing matches. You don't pick the mode — it's inferred from your request. :::

Who it's for

• Founders and solo builders who want to go from a one-line idea to a launched product without assembling a team.
• Engineering teams that want production-grade scaffolding — architecture, tests, security, CI/CD, observability, compliance — generated and enforced rather than hand-rolled.
• Anyone working inside Claude Code who wants to invoke a specific specialist (/drydock:security-engineer, /drydock:ux-designer, and so on) for a focused task.

What makes it different

Drydock is built so the output is trustworthy, not just plausible:

• Receipts — every agent writes JSON proof of its work, and gates verify the receipts (and the artifacts they claim) before opening.
• Re-anchoring — specs are re-read from disk at every phase transition, so long runs don't drift from the original requirements.
• Gates — production-ready is blocked on failing tests, coverage, performance budget, compliance controls, or architecture-boundary violations. An override requires a logged "accepted with justification" receipt.
• Security by default — a security-defaults protocol plus a real secret-guard hook that blocks secret writes/commits and scans staged diffs.
• Grounding — evidence-first generation: every claim cites file:line, command output, or a retrieved source, tagged [verified] / [inferred] / [unverified]. Agents never invent CVEs or CVSS scores.

tip

You only approve direction at the gates. Between them, the agents run autonomously — and the autonomy level you pick (Autopilot, Copilot, Checkpoint, or Manual) governs only the smaller, agent-level questions, never the three pipeline gates.

Next steps

• Installation — add the marketplace and install the plugin inside Claude Code.
• Quickstart — describe your first build and walk through the gates.