Claude Code TypeScript Starter
Give Claude Code projects versioned rules, workflows, specialist agents, and skills from the start.
- Source repo
- shinpr/ai-coding-project-boilerplate
- Stars
- ★ 240
- Last updated
- 1d ago
- License
- MIT
- Primary language
- JavaScript
- FA score
- 54/100 · Major gaps
At a glance
- How it runs
- Works with
- Platform-specificClaude Code
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A TypeScript team wants Claude Code rules, project conventions, and quality standards versioned alongside its code.
- Not a fit if
- Teams that do not use Claude Code
- Teams seeking a TypeScript starter without workflow setup
- Maintenance
- Commits in the last 30 dayslast commit 2026-11-20latest release v2.2.8 (2026-11-20)
- Source review
- 54/100 · Major gaps
What does this agent do, and when should you use it?
This starter kit creates TypeScript projects configured for Claude Code: create-ai-project adds a project-level CLAUDE.md, ready-to-use commands, specialized agents, and skills. New projects come with Biome formatting and linting and Vitest configured, plus a place to record context and quality standards for future sessions and contributors. Its commands cover clarification, design, planning, implementation, review, and diagnosis; applicable checks run during implementation, and design work can pause for approval. Lite Mode reduces some independent checks and agent calls. The setup lives in the project and runs through Claude Code, with Node.js 26.12.1+, pnpm, and Claude Code required.
Run npx create-ai-project my-project to generate a project and its Claude Code setup, install dependencies, then launch claude. In Claude Code, /project-inject records project context and /implement clarifies a request and inspects existing code; when needed, it prepares a design and plan for approval before implementing and running applicable checks. /design, /plan, and /build support staged handoffs, with /front-design, /front-plan, and /front-build for frontend work. /review checks completed work, while /diagnose investigates a problem without changing code. /create-skill and /refine-skill help create task-specific guidance. The updater can preview and replace managed Claude Code rules, commands, agents, and skills while preserving source code and existing package settings.
- A TypeScript team wants Claude Code rules, project conventions, and quality standards versioned alongside its code.
- A developer wants Claude Code to clarify a feature request, prepare a design and plan when needed, then implement and check the change.
- A team wants to approve a design first and continue planning or implementation later, possibly in another session or with another contributor.
- A maintainer wants to turn team knowledge for specific tasks into skills Claude can use when relevant.
- A project team wants fewer independent checks and agent calls while keeping code review and the usual quality checks.
Key setup facts
Pulled from the install and usage notes below. The project README has the current values.
- Install / run
npx create-ai-project my-projectnpx create-ai-project update --dry-runnpx create-ai-project update
- Runtime
- Node.js 26.12.1+
How do you install or deploy this agent?
Requires Node.js 26.12.1+, pnpm, and Claude Code. Create a project, install its dependencies, and launch Claude Code:
npx create-ai-project my-project
cd my-project
pnpm install
claudeAdd --lang=ja or --lang=zh-CN to the creation command for Japanese or Simplified Chinese workflow instructions.
How do you use this agent?
Start by adding project context, then ask for a change:
/project-inject
/implement Add rate limiting to the APIFor a staged design, plan, and implementation workflow:
/design Add rate limiting to the API
/plan
/buildSave Lite Mode as the project default:
node scripts/set-workflow-mode.js liteRestore Normal Mode:
node scripts/set-workflow-mode.js normalPreview and apply updates to managed setup files:
npx create-ai-project update --dry-run
npx create-ai-project update
claudeWhat are this agent's strengths and limitations?
- Generates a project-level
CLAUDE.md, commands, specialized agents, and skills that a team can maintain with its code. - Starts TypeScript projects with Biome and Vitest configured for formatting, linting, and tests.
- Offers both an end-to-end
/implementworkflow and staged design, planning, build, review, and diagnosis commands. - Provides Normal and Lite workflow modes; Lite reduces independent checks and agent calls.
- The workflows depend on Claude Code; teams using another coding assistant do not have a documented way to use these commands and agent definitions.
- Requires Node.js 26.12.1+, pnpm, and Claude Code, adding environment setup steps.
- The updater overwrites managed files it updates, so local customizations need to be committed and selectively restored afterward.
- The README does not state Claude Code costs or document offline use or support for other model platforms.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Claude Code TypeScript Starter This agent | 54 · Major gaps | CLI | ★ 240 | 1d ago | JavaScript | Claude Code |
| OpenCode Swarm | 53 · Major gaps | Agent plugin / skillFree + model costs | ★ 540 | 1d ago | TypeScript | — |
| GolemBot | 58 · Major gaps | CLIFree + model costs | ★ 324 | 21d ago | TypeScript | Codex · Claude Code |
| cc-skills-golang: Agent Skills for Production-Ready Go | 51 · Major gaps | Agent plugin / skillFree + model costs | ★ 3.9k | 12d ago | Go | ChatGPT · Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The README explains how Claude Code commands structure work, when approval is awaited, and which checks Lite mode omits. The frontend skills also require keeping secrets server-side and limiting sensitive log fields, supporting ordinary-use controls. Dependency security is only partly evidenced by version declarations and pinned CI actions; no audit or vulnerability response process is shown. External effects and recovery guidance are limited: updates overwrite managed files, while attribution includes author and repository fields but publisher identity is unverified.
The README's design, planning, implementation, and checking workflow generally aligns with the skills' document responsibilities and error-handling rules. package.json and CI specify Node, pnpm, and check tools, but the excerpts cannot establish ongoing dependency availability. CI explicitly leaves tests disabled, and the supplied material gives little evidence about runtime failure messages, so those areas receive partial scores.
The README targets TypeScript and Claude Code projects, gives scenarios from diagnosis through design, build, and review, and offers Japanese, English, and Simplified Chinese workflows plus Normal and Lite modes. Skills specify their triggers and technical prerequisites. The environment requirements are explicit, but evidence chiefly covers TypeScript, React, and Claude Code; support for other languages or agent environments is not established.
The packet includes language-organized guides, command examples, installation steps, stable naming, a package version, and a changelog entry point. The MIT license text is complete, justifying full marks for licensing. The README explains that updates overwrite managed files and advises committing custom changes first, providing limited limitations guidance. A complete FAQ, full changelog contents, and a clear maintenance commitment or update responsibility are not evidenced, so those criteria are not full marks.
The scaffold combines shared rules, specialized agents, skills, commands, and quality checks into a reusable workflow. Examples show practical paths from request to implementation and staged design, which can reduce setup and handoff effort. The packet does not quantify benefits or sufficiently address the added costs of multi-stage checks, documentation, and agent calls, so cost-benefit earns only partial credit.
The README makes specific claims about modes, approval points, and update behavior, with partial corroboration from package.json, CI, and skill excerpts. The supplied files are a sample, and the CI comment says tests are not enabled, limiting cross-source corroboration. Skills direct readers to distinguish observed facts from inference, and some claims are traceable, but these excerpts cannot verify every product claim.
- The CI test step remains commented out, and the packet contains no test execution evidence; treat stated quality checks as unverified.
- Updates overwrite managed rules, commands, agents, and skills. Commit custom content and review the dry-run and resulting diff before applying an update.
- Evidence for dependency security, runtime failure handling, and maintenance responsibility is limited in the supplied excerpts.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v2.2.8 Release: v2.2.8
Release notes excerpt
- Work Plan tasks take their verification method and Verification Focus from any governing source they cite, such as a UI Spec, not only the Design Doc. - Applied across English, Japanese, and Simplified Chinese.View on GitHub ↗ -
v2.2.7 Release: v2.2.7
Release notes excerpt
This patch makes agents find existing code by responsibility, gives implementers ownership of what they build, and routes problems they notice outside their change to review. Changes apply across English, Japanese, and Simplified Chinese. - Existing code is found by responsibility as well as by reference, so code that shares no name with the change can still be reused. A search that finds nothing is not treated as evidence of absence. …View on GitHub ↗ -
v2.2.6 Release: v2.2.6
Release notes excerpt
This patch makes post-implementation review read large changes in full and makes agents load the criteria they judge by. Changes apply across English, Japanese, and Simplified Chinese. - Reviews a Large change with one code-reviewer per unit of Work Plan tasks, at most 3 at a time. Medium changes keep one reviewer. - code-reviewer judges the changed code and tests directly instead of inventorying every acceptance criterion. The acceptanceCriteria output field is removed. …View on GitHub ↗
FAQ
Does using this starter kit cost money?
Can I use it with ChatGPT or another coding assistant?
Will the updater overwrite my project code?
Does Lite Mode skip code review?
Can I design a change now and implement it in another session?
/design stops for approval, and the approved design and plan documents can be used by a later session or another contributor.Is Claude Code TypeScript Starter open source?
Is Claude Code TypeScript Starter safe to use?
Is Claude Code TypeScript Starter still maintained?
Who makes Claude Code TypeScript Starter?
Thanks — we use this to improve the page.