From ship
Chains challenge, checkout, research, design, TDD implementation, quality, commit, and PR creation for end-to-end feature, refactoring, or migration work.
How this skill is triggered — by the user, by Claude, or both
Slash command
/ship:build [implementation task]When to use
一気通貫で実装, 新機能, 計画を要するリファクタリング, 大規模マイグレーション, build, end-to-end implementation
[implementation task]opusThis skill is limited to the following tools:
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
Chains `/challenge` → `/checkout` → `/research` → `/think` → `/code` → quality → `/commit` → `/pr`, running end-to-end from premise GO verification through committing test-verified implementation to PR creation. Covers not just new features but refactoring and migrations. Every phase is mandatory (no early exit).
Chains /challenge → /checkout → /research → /think → /code → quality → /commit → /pr, running end-to-end from premise GO verification through committing test-verified implementation to PR creation. Covers not just new features but refactoring and migrations. Every phase is mandatory (no early exit).
Take the implementation task from $ARGUMENTS (optional). If empty, ask the user what to implement via AskUserQuestion.
Run Skill("challenge", $ARGUMENTS) to verify the implementation premise as GO / NO-GO.
Run Skill("checkout", $ARGUMENTS) to create the working branch. Creating it after challenge GO avoids branching for a NO-GO proposal. The subsequent research / think / code artifacts and the final commit all land on this branch, keeping main clean. At the start the diff is empty, so naming leans mainly on $ARGUMENTS. Skip if already off the default branch.
Run Skill("research", $ARGUMENTS).
Output goes to .claude/workspace/research/YYYY-MM-DD-<slug>.md. Phase 5 detects it as prior research.
Run Skill("think", $ARGUMENTS). Generate SOW + Spec from the Phase 2 challenge output (verdict / decisions / trade-offs) and the Phase 4 research results as input.
Output is .claude/workspace/planning/YYYY-MM-DD-[feature]/sow.md + spec.md.
Run Skill("code", $ARGUMENTS). /code auto-detects the SOW from the Phase 5 output.
Get changed files with git diff main...HEAD --name-only. Phase 6 (/code) passes the mechanical baseline of lint / type / test / readability / coverage, so Phase 7 stacks two layers, internal audit and external polish. Audit runs a multi-dimensional review of security / resilience / causation / encapsulation under adversarial challenge with severity. Polish stacks external Codex (a non-Claude lens that counters Self-Enhancement bias) plus simplify / enhancer-code cleanup. Present any remaining issues to the user for a decision.
| Step | Action | Exit condition |
|---|---|---|
| 1 | Run Skill("audit") against the diff (assigns severity, adversarial challenge) | 0 critical/high → go to Step 3 |
| 2 | Fix each snapshot critical/high finding ID with Skill("fix", "<ID>"), then re-audit; this is one round | 0 critical/high or 3 rounds reached → go to Step 3 |
| 3 | Run Skill("polish") against the uncommitted diff (external Codex + cleanup, fixes directly) | Codex not installed → continue with cleanup only |
| 4 | Confirm tests pass | Test failure → fix (max 2). Still failing → present remaining issues |
If a polish fix breaks tests, polish stashes it internally. Surface any remaining issues before the Phase 8 commit.
After the quality layer and all tests pass, run Skill("commit", ...) to commit the changes in Conventional Commits format. Planning artifacts and implementation land in one commit. If issues remain unresolved, present them to the user before committing.
Run Skill("pr", $ARGUMENTS) to create a PR from the committed changes. pr previews the body and confirms creation via AskUserQuestion, so build adds no extra gate.
| Error | Action |
|---|---|
/challenge returns NO-GO | Present refuting evidence and stop |
/checkout finds branch exists | Propose an alternate name or continue on it |
/research or /think cancelled/failed | Save context and exit |
/code failed | Present the error and ask the user |
| Quality loop exhausted (3 rounds) | Present remainder, user decides |
/code quality gate shows AC unmet | Propose re-entry to Phase 6 or 7 |
/commit failed | Present the error and ask the user |
/pr failed/cancelled | Present the error and ask the user |
End only when every check holds. Present the reason to the user for any check that cannot be met.
/code AC coveragenpx claudepluginhub thkt/dotclaude --plugin shipOrchestrates a full build pipeline from PRD: plans tasks, implements in parallel, and reviews results. Routes simple work to a lighter workflow.
Implementation skill emphasizing verification-driven coding with tight feedback loops. Guides multi-step implementation work: orient, plan, implement, verify, commit. Based on analysis of 21k+ operations.
Executes scope-aware implementation with TDD and quality gates, scaling from inline plans for small changes to full multi-task workflows for larger work.