The 19-agent roster
Drydock is a team of 19 specialized agents coordinated by one orchestrator. You describe what you want in plain English; the orchestrator classifies the request, routes it to the right agents, and runs them through the pipeline — pausing only at the human approval gates.
Throughout these docs, "agent" and "skill" mean the same thing: a specialized worker the orchestrator routes to. Each is invocable as drydock:<name>.
:::info Pipeline phases A full build flows through six phases — DEFINE → BUILD → HARDEN → SHIP → LAUNCH → SUSTAIN — with three human approval gates: Gate 1 (requirements/BRD), Gate 2 (architecture), and Gate 3 (production readiness). See How it works for the full pipeline. :::
The full roster
The table below lists all 19 agents, what each owns, and the phase it primarily works in. The orchestrator routes to them automatically based on your request; you can also invoke any one directly.
| # | Agent | Owns | Phase |
|---|---|---|---|
| 1 | Orchestrator | Routing, gates, receipts | All (coordinator) |
| 2 | Polymath | Research, ideation, translation | DEFINE / Explore |
| 3 | Product Manager | Requirements (the WHAT) | DEFINE |
| 4 | Solution Architect | Architecture + API (the HOW) | DEFINE |
| 5 | UX Designer | UX research, IA, design-system spec | DEFINE → BUILD |
| 6 | Software Engineer | Backend services, APIs, data layers | BUILD |
| 7 | Frontend Engineer | Web UI, typed API clients, perf budgets | BUILD |
| 8 | QA Engineer | Test pyramid, coverage/mutation gates | HARDEN |
| 9 | Security Engineer | STRIDE/OWASP audit + VAPT | HARDEN |
| 10 | Code Reviewer | Code quality, architecture conformance | HARDEN |
| 11 | DevOps | CI/CD, IaC, containers, supply chain | SHIP |
| 12 | SRE | SLOs, runbooks, production-readiness review | SHIP |
| 13 | Data Scientist | LLM/ML optimization (conditional) | SHIP |
| 14 | Compliance Officer | Regulatory control mapping + gate | HARDEN |
| 15 | Technical Writer | API reference, developer/ops guides | SUSTAIN |
| 16 | Skill Maker | Project-specific Claude Code skills | SUSTAIN |
| 17 | Growth Marketer | Positioning, GTM/launch plan, marketing | LAUNCH |
| 18 | Sales Strategist | Pricing, collateral, sales process | LAUNCH |
| 19 | Customer Success | Onboarding, support, retention | LAUNCH → SUSTAIN |
:::note Not every request runs the whole roster The orchestrator picks an execution mode to match what you asked for. A code review or a single feature engages only the relevant agents, with fewer (or zero) gates. Only Full Build mode runs all 19. :::
Two ways agents run
The 19 agents split into two execution models. The distinction is about context isolation — where each agent runs and whether it can talk to you directly.
4 in-context agents
Four agents run in the main conversation as skills, sharing the orchestrator's context window:
- Orchestrator (
drydock) — classifies the request, builds the task graph, dispatches the other agents, and runs the approval gate ceremonies. - Product Manager — interviews you to produce the BRD; runs the Gate 1 conversation.
- Solution Architect — designs the system and API; runs the Gate 2 conversation.
- Polymath — research, ideation, and the gate "translator" that explains artifacts in plain language when you pick "Chat about this."
These run in-context because their work is sequential and user-interactive — they conduct interviews, surface decisions, and run approval gates that need the conversation with you. They are dispatched with the Skill tool:
Skill(skill="product-manager")
15 isolated subagents
The other 15 agents ship as subagent definitions at agents/<name>.md and run in their own isolated context window, backgrounded and parallel:
code-reviewer compliance-officer customer-success
data-scientist devops frontend-engineer
growth-marketer qa-engineer sales-strategist
security-engineer skill-maker software-engineer
sre technical-writer ux-designer
Each declares background: true in its own frontmatter and invokes the matching drydock:<name> skill in its body. The orchestrator delegates to them in natural language — it does not restate the role or pass subagent_type/isolation/background; those live in the subagent's own definition. They run their work autonomously, write a receipt JSON to drydock/.orchestrator/receipts/, and mark their task complete.
:::tip Why isolate? Isolated subagents do the heavy, autonomous work — implementing services, writing tests, running security audits — without polluting the main context window. That keeps the orchestrator's context lean across long runs and lets independent work proceed in parallel. :::
Worktree isolation
Most of the isolated subagents also declare isolation: worktree in their frontmatter. Each runs inside its own git worktree, so concurrent agents editing the codebase never clobber each other's files. After a parallel wave finishes, the orchestrator merges each worktree branch back into the working branch before the next phase reads those outputs.
A worktree subagent's worktree auto-cleans once its branch has been merged back — there is no team to tear down. On a gate rejection, in-flight subagents finish or are abandoned on their own and their worktrees auto-clean.
:::note Requirements Worktree isolation uses git. The build and ship phases also use Docker and Docker Compose (container builds, IaC). Install them if you plan to run a full build. :::
Invoking a single agent directly
You don't have to run the whole pipeline. Let the orchestrator route for you, or call any agent by name with /drydock:<name>:
/drydock:security-engineer audit my API for OWASP Top 10
/drydock:technical-writer document the public API
/drydock:code-reviewer review my latest changes
To let the orchestrator classify and route the request itself, invoke the orchestrator directly:
/drydock:drydock build a SaaS for booking dog walkers — auth, payments, dashboard
Every agent maps to a /drydock:<name> command — for example /drydock:product-manager, /drydock:solution-architect, /drydock:frontend-engineer, /drydock:devops, /drydock:growth-marketer. The name matches the agent's slug in the roster table.
The isolated subagents are designed to be dispatched by the orchestrator during a pipeline run. Invoking one directly still works, but the orchestrator normally sets up the workspace, re-anchors specs from disk, and verifies receipts at gates — context a standalone call skips.
Explore each category
The 19 agents group into five categories. Each page below covers the agents in that group, what they produce, and how they hand off to one another:
- Planning and design — Polymath, Product Manager, Solution Architect, UX Designer.
- Engineering — Software Engineer, Frontend Engineer, Data Scientist.
- Quality and security — QA Engineer, Security Engineer, Code Reviewer, Compliance Officer.
- Delivery — DevOps, SRE, Technical Writer, Skill Maker.
- Go-to-market — Growth Marketer, Sales Strategist, Customer Success.
Drydock is an open-source plugin for Claude Code (version 2.5.0). Install it from inside Claude Code:
/plugin marketplace add sundarshahi/drydock
/plugin install drydock@drydock
Source: github.com/sundarshahi/drydock