ship

Orchestrates a spec-driven delivery loop from approved plan through TDD, parallel review, and pull request.

Updated Sep 17, 2026
One-click install
npx skills add https://github.com/karenrebecag/spec-driven-standards --skill ship-karenrebecag
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: ship
Source: https://github.com/karenrebecag/spec-driven-standards/tree/main/plugins/standards/skills/ship
Command: npx skills add https://github.com/karenrebecag/spec-driven-standards --skill ship-karenrebecag

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Shipping code without enforced review gates leads to unreviewed commits, missed security issues, and inconsistent quality. This Skill runs a structured delivery loop where a commit cannot proceed without explicit APPROVE verdicts from code, security, and QA reviewers on the exact diff. ## Core Features & Use Cases - Spec-first planning: Launches a planner agent and stops for human approval of the spec before any code is written. - TDD enforcement: Requires a failing test before implementation, and every fix must start with a failing test rather than a direct patch. - Parallel mandatory review: Runs code-reviewer, security-reviewer, and qa-reviewer concurrently, with a hook-enforced gate that blocks commits lacking APPROVE verdicts. - Bounded fix rounds: Re-reviews only the reviewers that flagged issues, capped at 3 rounds before escalating to the human. - Use Case: You have a bug fix touching 3 files. Invoke /ship to get a spec approved, reproduce the bug with a failing test, implement the fix, pass all three reviewers, and end with an open pull request on a ship/ branch. ## Quick Start Invoke /ship with a concrete goal such as "fix the login timeout bug in the auth module" and approve the spec when the planner presents it.

Frequently Asked Questions about ship

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I enforce code review before commits in Claude Code?

Use the /ship loop, which runs code-reviewer, security-reviewer, and qa-reviewer in parallel and relies on a review-gate hook that blocks any commit lacking an APPROVE verdict for the exact diff. The gate reads each reviewer's VERDICT line and stores it against the diff hash.

How does the /ship skill enforce TDD?

The loop launches a tdd-guide agent that writes a failing test before any implementation, and every fix round must also start with a failing test rather than a direct patch. For QA findings, the fix is the missing test itself.

What happens if reviewers keep finding issues in /ship?

Fix rounds are capped at 3. Only the reviewers that flagged issues re-run on the new diff, and if CRITICAL or HIGH findings remain after the third round, the loop stops and escalates the pending findings to the human without committing.

Does /ship merge or deploy the pull request?

No. The loop ends at an open pull request created with gh pr create on a ship/ branch. Merging and deploying are explicitly reserved for the human operator.

What are the limitations of the /ship review gate?

The gate only covers commits made through the Bash tool by the agent; a manual git commit bypasses it. Work larger than 3-5 files must be split into separately mergeable phases by the planner first.

What dependencies does the qa-reviewer in /ship need?

The qa-reviewer drives the external awesome-qa-skills system, requiring at minimum the discover-testing router and the testing-types skills installed flat under ~/.claude/skills. Without them it falls back to the project's common testing rules.