Shipmates

Custom subagents & command workflows — for your AI coding harness.

Shipmates

Stop being your AI's for-loop. Give it a crew. 

One command takes a GitHub issue from open to a reviewed, CI-green pull request — autonomously.

Get the crew aboard View on GitHub
  • MIT licensed
  • 13 specialists
  • Zero config
A /shipmates-ship-issue run: Plan, Isolate, Build, Self-check, CI gate, Review, Remediate, Deliver — one GitHub issue driven to a reviewed, CI-green pull request.
Illustrative — the stages /shipmates-ship-issue runs, in order.

Come aboard

Get the crew aboard

One line, no clone. Install the CLI, then bring the crew aboard in any directory.

macOS / Linux (Homebrew)

brew install saman-mb/tap/shipmates

Anywhere (Cargo)

cargo install shipmates

Once installed, run shipmates install --harness <name> in the project directory to bring the crew aboard. See the full install reference for binary installers and per-harness paths.

Meet the crew

Thirteen specialist subagents

Domain-neutral specialists that work on anything — a game engine, a web app, a CLI. The bar they hold you to comes from your repo's README and CLAUDE.md, not the role.

  • architect

    Structure & schema — coupling, boundaries, migration safety.

  • senior-engineer

    Builds to spec, fixes red CI, clears review defects.

  • principal-engineer

    Principal diff review — quality, tests, repo ship checklist.

  • sdet

    Runs the real tests/build and reports pass/fail with a defect list.

  • security-engineer

    Threat-models the change — authz, injection, secrets, vulnerable deps.

  • site-reliability-engineer

    Reliability, failure modes, rollback safety — and bug root-cause.

  • performance-engineer

    Profiles, benchmarks, and proves the win.

  • devops-engineer

    Build & delivery — reproducibility, pinning, environment parity.

  • product-manager

    Accepts or rejects against the acceptance criteria and your bar.

  • ux-ui-designer

    Specs & reviews on-screen UI — tokens, layout, focus, a11y.

  • art-director

    Directs & reviews rendered visuals — judges the picture.

  • technical-writer

    Writes docs from the real code; proves them with a reader test.

  • data-scientist

    Data/model work — metrics, leakage, validation, reproducibility.

The commands

Command workflows

Reusable command workflows — each one convenes the right crew for a job and drives it to done. Open any command for its full stage-by-stage breakdown.

  • Flagship

    Ship Issue

    Drives issue #n from open → reviewed, CI-green PR with the whole crew.

  • Ship Epic

    Ships an epic's stories in dependency-ordered waves — independent units fan out concurrently; gates pause, failures pause with state.

  • Fix Bug

    Reproduce as a failing test first, root-cause, minimal fix, red→green proof.

  • Report Bug

    Files a structured bug on saman-mb/shipmates from a live run — preview by default, apply when ready.

  • Plan Epics

    Turns a brief into GitHub epics + linked, labelled user stories, in parallel.

  • Consolidate Issues

    Cross-checks open issues against history, merges duplicates into their survivors, closes stale work, and bundles the rest by theme.

  • Harden

    Threat-models a surface and ranks every finding — read-only by default, fixes opt-in.

  • Spike

    De-risks a decision — prototypes options in parallel, judges them, records an ADR.

  • Migrate

    Sweeps a migration across every call site, verified, no remnants left.

  • Document

    Writes docs from the real code, gated on a fresh reader completing the steps.

  • Release

    Cuts a release — changelog, CI-green tag, SRE rollback pre-flight, opt-in publish.

  • Polish

    Iterates a visual/UI artifact to a specialist's sign-off — render → critique → fix.

  • PR Review

    Runs the board against a PR the crew didn't author — reports, never repairs.

  • QA

    Interactive local human QA walkthrough — complements the board and CI.

  • Onboard

    Reads an unfamiliar repo and writes the agent-facing context file the crew needs.

  • Refactor

    Reshapes code without changing behaviour — characterization tests pinned first.

  • Deslop

    Audits the codebase for dead code, duplication and rot — risk-graded; fixes on a branch, opt-in.

  • Upgrade

    Checks for a newer Shipmates release, refreshes and audits every install — apply and file only on request.

  • Auto Trunk

    Runs a test-driven trunk loop — parse a goal, red→green each task, and commit to main. Plan mode is the default; commits and pushes are opt-in.

The tools

Tools the crew reaches for

A third kind of resource. A command is something you invoke with a slash; a tool is something the crew reach for on their own — implicitly, when the intent of your prompt calls for it. Never typed, never a slash command. Opt in at install with --with-tools.

  • shipmates-termgif

    Renders an animated terminal demo GIF of a workflow run from a small JSON spec.

  • shipmates-social-card

    Renders a 1280x640 social/OG share card from a small JSON spec.

  • shipmates-pixelart

    Renders pixel-art icons from a JSON spec as static PNG or animated GIF.

  • shipmates-diagram

    Renders a curated diagram — a flow/pipeline/state machine or a sequence of actors and messages — as a committed SVG, deterministic PNG, or animated GIF.

  • shipmates-svgflow

    Deprecated alias for shipmates-diagram — svgflow's flow diagram is now a kind of it.

  • shipmates-badge

    Render a shields-style status badge as an offline, committed SVG.

  • shipmates-sparkline

    Renders a short number series as a tiny inline SVG trend chart.

  • shipmates-scrub

    Redacts secrets and PII from a log or paste before it's shared.

  • shipmates-fixtures

    Generates deterministic fake test data from a small JSON schema.

  • shipmates-domaincheck

    Checks domain availability via RDAP — registry-authoritative verdicts, not DNS guesswork.

  • shipmates-gh

    Wraps the GitHub CLI with validated JSON ops — issues, PRs, CI checks, body-file hygiene.

How the voyage works

How /shipmates-ship-issue takes a GitHub issue to a pull request

It's not a clever prompt — it's a structured workflow with explicit quality gates. Eight stages, in order.

  1. Plan

    A planner reads the issue + your docs → build plan, acceptance criteria, validation plan, and which specialists the story needs.

  2. Design specs

    For UI / visual / architecture-heavy stories, the right specialist writes a spec the builders must build to. (Only if needed.)

  3. Isolate

    All work happens in a throwaway git worktree; your base branch never breaks.

  4. Build

    Parallel senior-engineer builders with non-overlapping file ownership.

  5. Self-check → CI gate

    The SDET runs the tests; then CI must go green on the pushed PR before anything moves on. Red? It reads the logs and fixes — bounded.

  6. Acceptance board

    product-manager + principal-engineer, plus gated specialists, review the pushed PR head adversarially.

  7. Remediate

    Any rejection loops back to a fixer, then re-selects the board from that delta — failers sit again, prior ACCEPTs carry unless the delta can invalidate them, and a seat the delta newly trips may join. Bounded, then escalates to you.

  8. Deliver

    Disposes non-blocking nits absorb-first (fixes cheap ones in the same PR; files only capped/batched leftovers), names a /shipmates-harden follow-up if the change touched a security-relevant surface, and opens (or, opt-in, merges) the PR.

What's next

Which harnesses the crew runs on

Claude Code is fully runtime-verified; opencode, antigravity, cursor, pi and codex have captain-attested live runs; GitHub Copilot, grok-build and devin remain format-only. Every claim is auditable against the harness matrix runtime_verified record rather than a promise.

  • Runtime-verified (full)
    • Claude Code — the full crew and all 19 commands; crew resolve, argument passing, and /shipmates-ship-issue end to end.
  • Live run (partial) — captain-attested live runs in tools/harness_matrix.json (#497; codex #510). Granular criteria stay unknown until separately logged, except pi crew_resolve and command_e2e which are no (#525).
    • opencode — full crew + all 19 commands as .opencode/; flagship tickets #31 and #32 are closed.
    • antigravity (the Antigravity CLI, agy — successor to the retired Gemini CLI) — full crew + all 19 commands as .agents/.
    • pi — full crew + all 19 commands at .pi/agents/ via pi-subagents; crew_resolve and command_e2e are no.
    • cursor — the 19 skills only (crew resolve N/A).
    • codex — full crew + all 19 commands; crew are TOML and inherit the session's tools (no per-agent allowlist).
  • Builds, not runtime-verified (none)
    • github-copilot — the full crew and all 19 commands; the crew-bearing target still without a live run (#41).
    • grok-build — the full crew and all 19 commands in its own .grok/ tree: the one non-Claude target that keeps the native disable-model-invocation guard, and it enforces a per-agent tool allowlist too.
    • devin — the full crew and all 19 commands in .devin/ (Devin CLI / Devin Desktop, formerly Windsurf), with the legacy .windsurf/ tree adopted on install.
  • Next
    • Promote partial rows to full by logging crew resolve, argument passing, and a named command E2E; a live run for GitHub Copilot (#41).

Why that's credible: the crew's system prompts name no harness, and the nineteen commands ship in the Agent Skills open-standard shape rather than a Claude-specific one — so most of a port is mapping frontmatter fields, not rewriting the crew. opencode is the first test of that: every persona and workflow body carried over unchanged, and what remained was path mapping and translating an allowlist into a permission map. Details on the harness matrix.

FAQ

Frequently asked questions

What is Shipmates?

A ready-made crew of subagents and command workflows. Instead of you playing planner–builder–reviewer in a loop, a board of specialist AI agents does it — the flagship /shipmates-ship-issue takes a GitHub issue all the way to a reviewed, CI-green pull request. It is harness-agnostic: one authored crew, compiled to whichever harness you install it for. See which harnesses for where each one stands.

What are subagents and skills?

Subagents are focused AI agents defined in .claude/agents/*.md; skills are reusable workflows defined in .claude/skills/<name>/SKILL.md and invoked as commands, like /shipmates-ship-issue. Shipmates ships 13 agents and 19 commands you drop into a repo's .claude/ with shipmates install (or .opencode/ for opencode, .codex/ for codex, and so on).

Is this an official Anthropic project?

No. Shipmates is an independent, MIT-licensed community project that builds on Claude Code's public subagent and skill features. “Claude” and “Claude Code” are trademarks of Anthropic.

How is it different from just prompting Claude Code?

A raw prompt drifts; Shipmates uses a structured workflow with explicit quality gates — an isolated worktree, a mandatory green-CI gate, and a fresh reviewer that never grades its own work — so an autonomous run converges instead of wandering.

Which languages and frameworks does it work with?

Any. The agents are domain-neutral — they enforce the standard in your repo's README / CLAUDE.md, so the same crew works on a game engine, a web app, or a CLI.

Do I have to configure each agent?

No. Install once, then /shipmates-ship-issue 42. The crew picks up your project's quality bar automatically from your repo's README / AGENTS.md.

Where are the full install docs?

The install reference covers every flag and path. See also how commands work, the harness matrix, troubleshooting, and architecture pages.