The Plugin

The Shipwright plugin (artifact A) is the core of the system. Install it once into any Claude Code session and it drives the full spec → plan → execute → review → deploy loop against any repository. It is pure TypeScript with no server, no database, and no external HTTP — everything runs locally inside Claude Code.

Installing the plugin

Ask Claude Code:

Install the Shipwright plugin so I can use it in this session.

Claude Code will run the install command and confirm when the plugin is active.

Manual install

Inside a Claude Code session:

/plugin install shipwright@app-vitals/shipwright

The plugin is available immediately — no restart required. It registers all commands, skills, and agent behaviors for the current session.

Commands

Commands are the hands-on tools for driving individual stages of the delivery loop. Each command is a slash command prefixed with /shipwright:.

Command	What it does
`/shipwright:prd`	Author or update a product requirements document
`/shipwright:plan-session`	Read the PRD and task queue, then produce a ready-to-execute task list
`/shipwright:dev-task`	Pick the next ready task and build it — tests first, then implementation, then open a PR
`/shipwright:review`	Deep single-pass code review of an open PR with inline comments
`/shipwright:patch`	Fix failing CI and unresolved review findings on open PRs
`/shipwright:deploy`	Merge an approved PR and watch post-merge CI
`/shipwright:metrics`	Pull pipeline metrics (cycle time, throughput, first-time quality)
`/shipwright:research`	Load relevant project docs and web research for the current task
`/shipwright:research-docs`	Refresh docs that have drifted from the code

Test-readiness pipeline

The five-phase test-readiness pipeline audits and improves your test coverage end to end:

Phase	Command	What it does
1	`/shipwright:test-inventory`	Catalogue existing tests and coverage gaps
2	`/shipwright:test-design`	Draft a test plan for uncovered areas
3	`/shipwright:test-migration`	Move tests to the right layer (unit / integration / e2e)
4	`/shipwright:test-roadmap`	Produce a prioritized remediation roadmap
5	`/shipwright:test-publish`	Publish the results to the task store

Run the full pipeline in one shot:

Run the full Shipwright test-readiness audit and publish results.

Claude Code will run each phase in order, wait for results, and publish the final report.

The tables above highlight the commands you’ll reach for most often — see the Commands Reference for the full catalog, including the maintenance patrol and planning/research commands not listed here.

Skills

Skills are autonomous behaviors that run multi-step loops without manual hand-holding.

Delivery loop

Skill	What it does
`pull-requests`	Query PR records from the task store — filter by repo, PR number, state, reviewState, or staged flag
`review-staged`	Walk through staged PR reviews from the task store conversationally — APPROVEs first, then COMMENTs, smallest diff to largest

Code health

Skill	What it does
`entropy-scan`	Scan codebase for golden principle deviations — report only, no code changes
`entropy-fix`	Read the entropy report, fix pr-worthy violations, and open targeted PRs — one PR per concern

Agent operations

Skill	What it does
`agent-admin`	Query and manage Shipwright agents via the admin API — cron jobs, env vars, tool permissions, API tokens, and plugins
`investigate-cron`	Diagnose why a cron run behaved unexpectedly — finds the session transcript by cron name and time
`learning-capture`	Capture a durable learning by editing CLAUDE.md or LEARNINGS.md directly
`task-store`	Query and update the Shipwright task store — pick the next ready task, mark status transitions, append new tasks

Test-readiness

Skill	What it does
`test-readiness`	Orchestrate the full test-readiness pipeline (phases 1–5) for the active worktree
`test-debt`	Compute a corrective-commit ratio per milestone to surface planning debt after a run

Plugin anatomy

The plugin lives at plugins/shipwright/ in the repo and is structured around four directories:

commands/ — every slash command listed above, as individual TypeScript files
skills/ — autonomous behavior scripts (agent-admin, entropy-scan, test-readiness, and more)
scripts/ — task-store adapters, check-docs-freshness.ts, and supporting tooling
references/ — schemas and recipes (metrics-schema.md, task-store.md, doc-refresh-recipe.md, reviews-schema.md)

The plugin is repo-agnostic: it carries no opinions about your stack, reads your test command from context, and adapts. It never hard-codes paths, frameworks, or CI providers.

Design constraints

No new coupling — the plugin does not import the agent, the metrics service, or the task store at runtime.
No server, no database — all plugin execution happens inside the Claude Code process.
Local-first — works offline against any local git repository. No cloud account required.

Next steps

With the plugin installed, you can run the full local stack — including the metrics dashboard and dev agent — using task stack. See Running Locally for the full setup.

← Previous Next →