By myWsq
计划驱动开发三件套:dev-explore 摸清代码地形、澄清需求并收敛方案,dev-write-plan 写成自包含实现计划,dev-execute-plan 在当前分支落地——默认主 agent 自己实现,也可委派 codex / cursor 实现再由主 agent 评审。三者可单独使用,顺跑时链式交接。
Execute an implementation plan written by dev-write-plan under `plans/` on the current branch. Use when the user asks to implement, run, execute, or delegate a plan such as `plans/001`, `execute 002`, `use codex/cursor/claude for this plan`, or `run the next TODO plan`. Supports self-execution or delegation to detected agent CLIs, then verifies and reviews the diff.
Read-only codebase and requirement exploration before planning. Use when the user asks how code works, wants relevant files and validation commands identified, has an unclear or open-ended development request, or needs implementation approaches compared before dev-write-plan. Never edits files; reports findings, clarified requirements, and an approved design direction in chat.
Turn a clear development request into a self-contained implementation plan under `plans/`, suitable for dev-execute-plan or another agent to execute. Use when the user asks to plan, design implementation steps, convert a bug/feature request into an executable plan, or continue after dev-explore. Planning is read-only except for files under `plans/`.
简体中文 | English
dev-skills is a small collection of agent skills for plan-driven software development. It separates a development task into three explicit phases: code exploration, implementation planning, and plan execution. The goal is to help an agent build enough context before coding, produce a reviewable plan before implementation, and finish with validation commands plus a git diff review.
It is also designed for cross-agent collaboration: use a more capable agent to explore and write the plan, then use a faster or cheaper agent to implement that plan, with the lead agent reviewing the final diff.
This repository supports three distribution paths:
dev plugin from a Codex plugin marketplace;dev plugin from a Claude Code plugin marketplace;npx skills when you only need part of the workflow.| Skill | Purpose | Output |
|---|---|---|
dev-explore | Read the relevant code, identify current behavior, validation commands, conventions, ambiguous requirements, and viable approaches without modifying files. | A codebase map, requirement clarification, and approved direction in the conversation. |
dev-write-plan | Convert a clear requirement into a self-contained, executable, and verifiable implementation plan. | plans/NNN-*.md plus the plans/README.md index. |
dev-execute-plan | Execute a plan on the current branch, validate it, and commit the work; it can also delegate implementation to a supported local agent and review the resulting diff. | Implementation commits and plan status updates on the current branch. |
The skills can be used independently, but they are designed to work well as a sequence:
dev-explore -> dev-write-plan -> dev-execute-plan
codex plugin marketplace add myWsq/dev-skills
codex plugin add dev@dev-skills
This installs a plugin named dev that includes all three skills in this repository.
/plugin marketplace add myWsq/dev-skills
/plugin install dev@dev-skills
This installs a plugin named dev that includes all three skills in this repository.
npx skillsUse this path when you only want one skill, or when your agent consumes plain SKILL.md directories directly.
Install all skills:
npx skills add myWsq/dev-skills
Install a single skill:
npx skills add myWsq/dev-skills --skill dev-explore
Useful options:
--list: list the skills available in this repository.-g: install globally for reuse across projects.Use dev-explore when the requirement is still unclear, the relevant code path is uncertain, or you need to understand the existing architecture and validation workflow first. For open-ended or behavior-changing requests, it also compares approaches and converges on a direction before planning, preferably using structured question tools when the environment provides them. This skill is read-only: it does not create plan files or modify source code.
Use dev-write-plan when the requirement is clear enough to turn into executable steps for an agent. The generated plan includes scope, steps, validation commands, completion criteria, and stop conditions.
Use dev-execute-plan when an implementation plan already exists under plans/ and should be applied on the current branch. It checks the working tree, follows the plan, and uses commits plus diffs as review boundaries.
Recommended workflow:
smart agent: dev-explore -> dev-write-plan
fast/cheap agent: dev-execute-plan
lead agent: review the diff and validation results
dev-execute-plan can execute the plan directly, or delegate implementation to a detected local agent CLI:
| Agent | How it is used |
|---|---|
codex | Runs codex exec in the current repository. |
cursor | Runs cursor-agent in the current repository. |
claude | Runs Claude Code in headless mode in the current repository. |
Direct execution by the current agent is always available. Delegation depends on which agent CLIs are installed and authenticated on your machine.
Example prompts:
Use dev-explore to understand how authentication works in this repo.
Use dev-explore to inspect the billing flow before we plan changes.
Use dev-write-plan to plan adding password reset support.
Use dev-write-plan to turn this bug report into an implementation plan.
Use dev-execute-plan to implement plans/001.
Use dev-execute-plan to execute the next TODO plan.
Use dev-execute-plan to delegate plans/002 to a faster local agent and review the result.
MIT
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub mywsq/dev-skills --plugin devDesign fluency for frontend development. 1 skill with 23 commands (/impeccable polish, /impeccable audit, /impeccable critique, etc.) and curated anti-pattern detection.
Behavioral guidelines to reduce common LLM coding mistakes, derived from Andrej Karpathy's observations on LLM coding pitfalls
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Access thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.
Develop, test, build, and deploy Godot 4.x games with Claude Code. Includes GdUnit4 testing, web/desktop exports, CI/CD pipelines, and deployment to Vercel/GitHub Pages/itch.io.
Complete developer toolkit for Claude Code