shinpr/claude-code-workflows

Claude Code Development Workflows

What it is – A collection of Claude Code plugins that turn the Claude Code AI‑assistant into a structured development‑process engine. The plugins encode a set of “recipes” (e.g. /recipe‑implement, /recipe‑design, /recipe‑review) that guide the model through:

1. Scope agreement – the user and Claude agree on the desired outcome and any exclusions.
2. Design & planning – Claude generates design docs, UI specs, ADRs and a work‑plan that is reviewed before any code is written.
3. Task‑level implementation – each work‑plan item is executed, verified against repository‑specific quality rules, and committed automatically.
4. Independent review – a second‑stage security/review agent checks the whole change for functional, reliability or security regressions before the workflow finishes.

The system is deliberately split into three plugin families:

Plugin	Target	Typical entry point
dev‑workflows	Backend or generic codebases	/recipe‑implement, /recipe‑design, …
dev‑workflows‑frontend	React/TypeScript front‑ends	/recipe‑front‑design, /recipe‑front‑build, …
dev‑workflows‑fullstack	Projects that have both back‑ and front‑end parts	/recipe‑fullstack‑implement, …

Why you might use it

• You need a formal hand‑off between a design discussion and actual code generation so that Claude doesn’t drift away from the agreed product goal.
• Your team wants repeatable safety gates (design review, work‑plan approval, independent security review) that are enforced automatically by the plugin.
• You already use Claude Code and would like to keep the same AI‑driven workflow across many repositories without reinventing the process each time.

Quick start (from the README)

# 1. Start Claude Code
claude

# 2. Add the marketplace that hosts the plugins
/plugin marketplace add shinpr/claude-code-workflows

# 3. Install the plugin that matches your project type
#   – backend / generic
/plugin install dev-workflows@claude-code-workflows
/recipe-implement "Add rate limiting to the public API"

#   – React frontend
/plugin install dev-workflows-frontend@claude-code-workflows
/recipe-front-design "Add account recovery screens"

#   – full‑stack
/plugin install dev-workflows-fullstack@claude-code-workflows
/recipe-fullstack-implement "Add user authentication with JWT + login form"

*Only one workflow plugin should be installed per repository; the full‑stack plugin already bundles the backend and frontend recipes.

How the workflow works (high‑level flow)

Request → Agree on outcome & exclusions →
   ┌─► Simple path? ──► Direct task cycle → Complete
   │
   └─► Complex? → Inspect → Design → Review → Approve scope →
        per‑task: implement → verify → quality‑check → commit →
        independent security review → (corrections loop) → Complete

The decision of which path to take depends on the complexity of the product decision, not on file count.

Core concepts introduced by the plugins

• Work Plan – a table that maps every approved requirement (from a Design Doc, ADR, UI Spec, etc.) to one or more concrete tasks. Gaps are flagged so nothing is left undocumented.
• Task template – each generated task carries the exact acceptance criteria and a yes/no compliance check that runs before the commit.
• Quality profile – repository‑specific linting, test, and security rules stored in docs/project‑context/quality.yaml. The plugin reads this profile during every verification step.
• Specialised agents – the plugin orchestrates a suite of AI agents (e.g. requirement‑analyzer, code‑verifier, security‑reviewer) that keep design, implementation and review concerns separate.

Example run (summarised from the README)

1. User runs /recipe‑implement "Add rate limiting to the public API".
2. Claude analyses the repo, proposes a design doc, and the user approves it.
3. A work‑plan is generated, split into 4 tasks (add middleware, update config, write tests, update docs).
4. For each task the task‑executor agent writes code, the code‑verifier runs the repository’s quality profile, and the change is committed.
5. After all tasks finish, the security‑reviewer agent runs an independent check; it finds no issues, so the workflow ends.

When it’s not needed

If the change is a quick throw‑away prototype or the outcome and safe implementation boundaries are already crystal‑clear, you can just use Claude Code directly without the extra workflow overhead.

---

License & contribution

• License: MIT (see badge in README).
• Contributions: PRs are welcome; the repository includes a PRs welcome badge.

---

Bottom line

shinpr/claude-code-workflows is not a library that you import into your own code; it is a Claude Code plugin suite that adds disciplined, AI‑driven development pipelines to any project that uses Claude Code. By codifying design‑review, work‑plan approval, task‑level verification and independent security review, it helps teams keep Claude‑generated code aligned with agreed product goals and repository quality standards.