using-shipyard

Using Shipyard

What is Shipyard?

Shipyard is a structured project execution framework for Claude Code. It works as a plugin that helps you:

• Plan work in phases — break large projects into manageable pieces with clear success criteria
• Build with parallel agents and TDD — fresh 200k-token context per task, atomic commits, test-driven development
• Review code quality and security automatically — two-stage code review, OWASP security audits, complexity analysis
• Ship with confidence — verification gates, documentation generation, and delivery workflows

Shipyard works as a Claude Code plugin. Install it once, then use slash commands in any project.

Getting Started

New to Shipyard? Follow these steps:

1. /shipyard:init — Set up your project preferences (interaction mode, git strategy, quality gates). Takes ~1 minute.
2. /shipyard:brainstorm — Explore what you want to build through interactive dialogue. Captures a project definition.
3. /shipyard:plan 1 — Plan your first phase of work. Researches the codebase and decomposes into executable tasks.
4. /shipyard:build — Execute the plan with parallel builder agents, review gates, and security audits.
5. /shipyard:ship — Verify, audit, document, and deliver your completed work.

For quick one-off tasks, skip the lifecycle and use /shipyard:quick 'description'.

I Want To...

Goal	Command
Set up a new project	/shipyard:init
Explore requirements	/shipyard:brainstorm
Understand existing code	/shipyard:map
Plan a phase	/shipyard:plan [phase] [--skip-research]
Build from a plan	/shipyard:build [phase] [--plan N] [--light]
Quick one-off task	/shipyard:quick "task"
Review my code	/shipyard:review
Security check	/shipyard:audit
Research technology options	/shipyard:research "topic"
Find duplication/complexity	/shipyard:simplify [scope]
Generate documentation	/shipyard:document [scope]
Run tests and verify	/shipyard:verify [criteria]
Check progress	/shipyard:status
Change settings	/shipyard:settings
Rollback to checkpoint	/shipyard:rollback
Recover from errors	/shipyard:recover
Ship completed work	/shipyard:ship [--phase | --milestone | --branch]

How to Access Skills

In Claude Code: Use the Skill tool. When you invoke a skill, its content is loaded and presented to you — follow it directly. Never use the Read tool on skill files.

In other environments: Check your platform's documentation for how skills are loaded.

Available Skills

Shipyard provides 19 skills. Skills are behavioral disciplines — they define HOW to do work, not what to build.

Skill	What It Actually Does
shipyard:using-shipyard	Index of all skills/commands with triggers and activation rules (this skill)
shipyard:shipyard-tdd	Enforces write-failing-test → watch-fail → implement → watch-pass → refactor cycle. Tests written after code prove nothing.
shipyard:shipyard-debugging	4-phase investigation (root cause → pattern analysis → hypothesis test → fix). After 3 failed fixes, stop and question architecture.
shipyard:shipyard-verification	Blocks "done"/"fixed"/"passing" claims until you run the actual command fresh and read the output. Evidence before assertions.
shipyard:shipyard-brainstorming	One-question-at-a-time Socratic dialogue to turn vague ideas into validated designs with trade-offs before any code is written.
shipyard:security-audit	Checks OWASP Top 10, scans for hardcoded secrets, audits dependencies and IaC (Terraform/Ansible/Docker). Critical findings block merge.
shipyard:code-simplification	Post-implementation review for duplication (3+ occurrences → extract), dead code, over-engineering, and AI bloat patterns.
shipyard:infrastructure-validation	Tool-chain workflows for Terraform (fmt/validate/plan/lint), Ansible (lint/syntax/dry-run), Docker (hadolint/build/trivy).
shipyard:parallel-dispatch	Routes 2+ independent tasks to concurrent agents with isolated scope/constraints. Prevents sequential bottleneck.
shipyard:shipyard-writing-plans	Converts specs into executable tasks with exact file paths, code samples, verification commands, and TDD steps for builder handoff.
shipyard:shipyard-executing-plans	Runs fresh builder agent per task + two-stage review (spec compliance then code quality) + security audit + simplification gate.
shipyard:git-workflow	Full branch lifecycle: create worktree → setup → baseline tests → work → verify → 4 completion options (merge/PR/keep/discard).
shipyard:documentation	Generates code comments, API docs (params/returns/examples), architecture docs, and user guides. Verifies examples actually work.
shipyard:shipyard-writing-skills	TDD for docs: run scenario WITHOUT skill (RED) → document agent rationalizations → write skill countering them (GREEN) → refine.
shipyard:shipyard-testing	Enforces behavior-based testing via public APIs: AAA structure, DAMP not DRY, name tests after behavior, prefer state over mocks.
shipyard:lessons-learned	After phase completion, captures what worked/surprised/failed into .shipyard/LESSONS.md and optionally feeds back to CLAUDE.md.
shipyard:shipyard-handoff	Captures session context into .shipyard/HANDOFF.md so the next session can resume without losing progress.
shipyard:import-spec	Imports a spec-kit feature directory into Shipyard, replacing brainstorming. Maps spec artifacts to PROJECT.md and ROADMAP.md.
shipyard:import-spec-file	Imports a handwritten spec document into Shipyard, replacing brainstorming. Interviews to fill gaps via Socratic dialogue.

Shipyard Commands

Commands are actions — they produce artifacts, change state, or trigger workflows.

Lifecycle Commands (run in order for full projects)

Command	What It Actually Does
/shipyard:init	Settings only. Asks ~9 preference questions (interaction mode, git strategy, review depth, quality gates, model routing) and writes .shipyard/config.json. No codebase analysis or planning.
/shipyard:brainstorm	Socratic dialogue exploring goals/constraints → writes .shipyard/PROJECT.md. Optionally dispatches architect to generate ROADMAP.md with up to 3 revision cycles.
/shipyard:plan [phase] [--skip-research]	Dispatches researcher → architect → verifier agents to create executable plan files (PLAN-W.P.md) with tasks, file paths, verification commands. Produces RESEARCH.md and context files.
/shipyard:build [phase] [--plan N] [--light]	Dispatches builder agent per plan + two-stage review + optional security audit/simplification/docs. Handles retries (up to 2) on critical issues. Produces SUMMARY and REVIEW files per plan.
/shipyard:ship [--phase | --milestone | --branch]	Pre-ship verification + tests + security audit + docs + lessons learned capture. Then presents 4 delivery options: merge locally, push PR, preserve branch, or discard. Archives artifacts.

Session Management

Command	What It Actually Does
/shipyard:status	Reads state files and displays progress dashboard with blockers and next-action suggestion. No side effects.
/shipyard:resume	Reconstructs context from STATE.json + HISTORY.md + artifacts to detect interrupted work and route to recovery.
/shipyard:settings	Interactive menu to view/update .shipyard/config.json preferences. Supports list, view <key>, set <key> <value>.
/shipyard:issues	Manages deferred issues in .shipyard/ISSUES.md with severity tracking. Supports --add, --resolve, --list.
/shipyard:rollback	Reverts to a git checkpoint (state-only or full code+state). Lists recent checkpoints, creates safety checkpoint before reverting.
/shipyard:recover	Diagnoses state inconsistencies and offers recovery: resume, rollback, rebuild state from artifacts, or full reset.

On-Demand Tools (use anytime, no lifecycle required)

Command	What It Actually Does
/shipyard:quick "task"	Dispatches architect (simplified plan) → builder (execute + verify) for small self-contained tasks. Produces QUICK-NNN.md. No roadmap needed.
/shipyard:review [target]	Two-stage code review (spec compliance if plan exists, otherwise quality). Accepts uncommitted changes, diff ranges, file paths, or branch comparisons.
/shipyard:audit [scope]	OWASP Top 10 + secrets + dependencies + IaC security scan. Critical findings are hard gates. Accepts changes, ranges, directories, or full codebase.
/shipyard:simplify [scope]	Detects duplication, dead code, unnecessary abstractions, and AI bloat. Non-blocking findings. Accepts changes, ranges, or directories.
/shipyard:document [scope]	Generates API docs, architecture updates, user guides. Analyzes code vs existing docs to find gaps. Accepts changes, ranges, or directories.
/shipyard:research <topic>	Dispatches researcher to evaluate technology options via web search + codebase analysis. Produces comparison matrix with recommendations.
/shipyard:verify [criteria]	Runs acceptance criteria or test suites, records PASS/FAIL with evidence. Accepts test suite, criteria file, phase number, or inline text.
/shipyard:debug <error>	Dispatches debugger agent for systematic root-cause analysis using 5 Whys protocol. Accepts error descriptions or test names.
/shipyard:map [focus]	Deep codebase analysis producing up to 7 docs (STACK, INTEGRATIONS, ARCHITECTURE, STRUCTURE, CONVENTIONS, TESTING, CONCERNS). Supports parallel "all" mode.
/shipyard:worktree	Git worktree management: create (isolated branch + setup + tests), list, switch, remove (with dirty-state safety checks).
/shipyard:move-docs	Relocates codebase docs between .shipyard/codebase/ (private) and docs/codebase/ (public) using git mv.

Skill Activation Protocol

When a trigger condition matches, invoke the corresponding skill before responding.

File Pattern Triggers

Pattern	Skill
*.tf, *.tfvars, terraform*	shipyard:infrastructure-validation
Dockerfile, docker-compose.yml, *.dockerfile	shipyard:infrastructure-validation
playbook*.yml, roles/, inventory/, ansible*	shipyard:infrastructure-validation
*.test.*, *.spec.*, __tests__/, *_test.go	shipyard:shipyard-tdd

Task Marker Triggers

Marker	Skill
tdd="true" in plan task	shipyard:shipyard-tdd
Plan file loaded for execution	shipyard:shipyard-executing-plans
Design discussion, feature exploration	shipyard:shipyard-brainstorming
Creating an implementation plan	shipyard:shipyard-writing-plans

State Condition Triggers

Condition	Skill
About to claim "done", "complete", "fixed"	shipyard:shipyard-verification
About to commit, create PR, or merge	shipyard:shipyard-verification
Bug, error, test failure, unexpected behavior	shipyard:shipyard-debugging
2+ independent tasks with no shared state	shipyard:parallel-dispatch
Creating or editing a skill file	shipyard:shipyard-writing-skills
Branch management, delivery, worktrees	shipyard:git-workflow
Starting feature work on a new phase	shipyard:git-workflow
SHIPYARD_IS_TEAMMATE=true in env	shipyard:shipyard-executing-plans, shipyard:shipyard-verification — follow teammate mode sections

Content Pattern Triggers

Pattern in output or conversation	Skill
Error, exception, traceback, failure	shipyard:shipyard-debugging
Security, vulnerability, CVE, OWASP	shipyard:security-audit
Duplicate, complex, bloat, refactor	shipyard:code-simplification
Document, README, API docs, changelog	shipyard:documentation

Natural Language Triggers

Phrase	Skill
"what should I do", "help me", "how do I", "where do I start"	shipyard:using-shipyard
"build this", "implement this", "execute the plan"	shipyard:shipyard-executing-plans
"fix this bug", "why is this failing", "debug this", "it's broken"	shipyard:shipyard-debugging
"test first", "write tests", "TDD", "red green refactor"	shipyard:shipyard-tdd
"I want to add", "let's design", "brainstorm", "what if we"	shipyard:shipyard-brainstorming
"is this done", "verify this", "check my work"	shipyard:shipyard-verification
"add tests", "test coverage", "testing strategy"	shipyard:shipyard-testing
"security check", "is this secure", "vulnerability scan"	shipyard:security-audit
"simplify this", "clean up", "too complex"	shipyard:code-simplification
"document this", "write docs", "update README"	shipyard:documentation
"create branch", "commit this", "merge branch", "start feature"	shipyard:git-workflow
"what did we learn", "capture lessons", "retrospective"	shipyard:lessons-learned
"validate terraform", "check docker", "lint ansible"	shipyard:infrastructure-validation
"run in parallel", "do these at the same time"	shipyard:parallel-dispatch
"write a plan", "create a plan", "plan this feature"	shipyard:shipyard-writing-plans
"create a skill", "write a skill", "new skill"	shipyard:shipyard-writing-skills
"handoff", "hand off", "transfer session", "save context", "I'm done for now"	shipyard:shipyard-handoff
"import spec", "import my spec", "use this spec"	shipyard:import-spec, shipyard:import-spec-file

The Core Rule

Invoke relevant skills BEFORE any response or action. If there is a reasonable chance a skill applies, invoke it to check. If an invoked skill turns out to be wrong for the situation, you don't need to use it.

Before every response, evaluate triggers in this order:

1. File patterns — check files being discussed, modified, or created
2. Task markers — check any loaded plans or task definitions
3. State conditions — check current workflow state and intent
4. Content patterns — check recent output and user messages

If any trigger matches, invoke the skill before responding. Multiple triggers can fire simultaneously.

Conflict Resolution

When multiple skills could activate for the same situation, follow this priority chain:

1. Debugging (shipyard-debugging) — always investigate root cause before anything else
2. TDD (shipyard-tdd) — if writing code, tests come first
3. Verification (shipyard-verification) — before claiming anything is done
4. Brainstorming (shipyard-brainstorming) — design before implementation
5. Security (security-audit) — security concerns override feature work
6. All others — apply in the order they match

Example: User says "fix this bug and add tests" → debugging first (investigate), then TDD (write failing test), then verification (prove it works).

Red Flags

These thoughts indicate a missed skill invocation:

Thought	What to do
"This is just a simple question"	Questions are tasks. Check for skills.
"Let me explore the codebase first"	Skills tell you HOW to explore. Check first.
"This doesn't need a formal skill"	If a skill exists for it, use it.
"I'll just do this one thing first"	Check BEFORE doing anything.

Skill Priority

When multiple skills could apply: process skills first (brainstorming, debugging), then implementation skills (executing-plans, parallel-dispatch).

Skill Types

Rigid (TDD, debugging, verification): Follow exactly. Flexible (patterns): Adapt to context. The skill itself tells you which.

User Instructions

Instructions say WHAT, not HOW. "Add X" or "Fix Y" doesn't mean skip workflows.

Good: Skill invoked before responding

User: "Add a health check endpoint"

Before responding, check triggers:

• Task marker → this is implementation → check for shipyard:shipyard-tdd (writing code)
• State condition → starting feature work → check for shipyard:git-workflow (branch management)

Result: Invoke shipyard:shipyard-tdd, then proceed with implementation using TDD discipline.

Bad: Skipping skill invocation

User: "Add a health check endpoint"

Immediately start writing code without checking triggers. Miss TDD discipline, skip branch creation, no verification before claiming done.

Good: Multiple triggers fire

User: "Fix the SQL injection bug in the auth module"

Triggers matched:

1. Content pattern → "bug" → invoke shipyard:shipyard-debugging (investigate root cause first)
2. Content pattern → "SQL injection" → invoke shipyard:security-audit (check for related vulnerabilities)
3. State condition → will claim "fixed" when done → invoke shipyard:shipyard-verification (verify before claiming)

Result: Debug first, audit for related issues, verify the fix works.