21 battle-tested methods
Not prompt templates — workflow methodology hardened by cross-session adversarial testing, from brainstorming to TDD, debugging and review.
superpowers 300k+ ⭐ · Full CN localization + China-native skills
Give your AI coding tools
superpowers that actually ship
21 battle-tested workflow skills — from brainstorming to TDD, systematic debugging to code review.
One command auto-detects your tool and installs.
$ npx superpowers-zh21Skills
4China-native
26Tools
v1.8.0Version
Why superpowers-zh?
Not another prompt-template pack — real engineering methodology that makes AI work properly.
Works in 26 tools
One skill set for Claude Code / Cursor / Codex / Gemini CLI / Windsurf and more. Switch tools, keep your habits.
One-command install
npx superpowers-zh auto-detects your tool and installs. Zero config; restart to take effect.
China-native skills
Chinese code-review phrasing, commit conventions, doc typography, and domestic Git platforms (Gitee/Coding/JiHu) — not in upstream.
Fully localized upstream
Tracks obra/superpowers (300k+ ⭐); every core skill localized into native Chinese — calibrated, not machine-translated.
Zero-dep · MIT
Pure Markdown skills. No external deps, no network, no code upload. Triggered on demand with zero runtime cost.
An end-to-end workflow, every step guarded by a skill
Skills chain together; the AI triggers the right methodology at the right moment.
1
BrainstormClarify intent before coding
→
2
PlanBreak work into steps
→
3
TDDTest first, then implement
→
4
DebugReproduce & locate first
→
5
ReviewRigorous pre-merge check
→
6
VerifyProve it with evidence
After install, AI stops "coding before thinking"
❌ Without
You: Add bulk export to the users module
AI: Sure, implementing… (starts coding)
export async function exportUsers() { … }
You: Wait — wrong format, no paging,
it OOMs on large data…✅ With superpowers-zh
You: Add bulk export to the users module
AI: (triggers brainstorming) One question:
who exported user data last time, and why?
You: Ops, weekly, ~300k rows, for BI
AI: Here is what I understand — confirm first
→ plan → TDD → verify → reviewPick your tool, get the command
For most tools npx superpowers-zh auto-detects the project and installs; otherwise pass --tool.
npx superpowers-zh21 skills, covering the whole dev workflow
Click any card for the full operating doc.
Using Superpowers · Bootstrap
using-superpowers
The bootstrap skill — establishes how to discover and invoke skills at the start of every conversation.
Read docs →Brainstorming
brainstorming
Find out what your human partner actually wants, and why, before building anything — features, behavior changes, or new work outside software.
Read docs →Dispatching Parallel Agents
dispatching-parallel-agents
Fan out 2+ independent tasks with no shared state or ordering dependency.
Read docs →Executing Plans
executing-plans
Execute a plan yourself in the current session — inline, with a ledger and one final review.
Read docs →Subagent-Driven Dev
subagent-driven-development
Run a plan of independent tasks within the current session via subagents.
Read docs →Workflow Runner
workflow-runner
Run agency-orchestrator YAML workflows directly using the current session LLM — no API key.
Read docs →Writing Plans
writing-plans
Turn a spec into a step-by-step implementation plan before writing any code.
Read docs →Diagnosing Superpowers
diagnosing-superpowers
When a session went wrong, read its transcripts and report what happened with path:line evidence — optionally as an issue.
Read docs →Systematic Debugging
systematic-debugging
Reproduce and locate the root cause before proposing any fix.
Read docs →Test-Driven Development
test-driven-development
Write the test before the implementation, for every feature and bug fix.
Read docs →Verification Before Completion
verification-before-completion
Run verification and back every claim with evidence before saying it is done.
Read docs →Receiving Code Review
receiving-code-review
Apply review feedback with technical rigor — verify, don't blindly comply.
Read docs →Requesting Code Review
requesting-code-review
Validate work against requirements before merging or shipping.
Read docs →Finishing a Branch
finishing-a-development-branch
Wrap up finished work via structured merge / PR / cleanup options.
Read docs →Using Git Worktrees
using-git-worktrees
Start isolated feature work in a dedicated git worktree.
Read docs →MCP Builder
mcp-builder
Methodology for building production-grade MCP servers that connect AI to external tools.
Read docs →Writing Skills
writing-skills
Create, edit and validate skills before deploying them.
Read docs →Chinese Code Review
China-nativechinese-code-review
Chinese review phrasing, severity tiers, and common anti-patterns in domestic teams.
Read docs →Chinese Commit Conventions
China-nativechinese-commit-conventions
Conventional Commits adapted for Chinese, with commitlint / husky / changelog templates.
Read docs →Chinese Documentation
China-nativechinese-documentation
CN/EN spacing, punctuation, term preservation and Chinese typography conventions.
Read docs →Chinese Git Workflow
China-nativechinese-git-workflow
Gitee / Coding.net / JiHu GitLab / CNB access, credentials, CI and mirror sync.
Read docs →No matching skills.
Typical use cases
Each scenario is backed by a set of cooperating skills.
brainstorming → writing-plans → TDDAI questions the requirements, writes a plan, then builds test-first — instead of dumping code immediately.
systematic-debuggingForces reproduce-and-locate-root-cause before any fix — no more "guess and tweak" loops.
verification-before-completion → code-reviewMust run verification with evidence, then a review pass, before claiming "done".
chinese-commit-conventions → chinese-code-reviewChinese commit conventions + tiered review phrasing, wired for Gitee / Coding / JiHu GitLab.
One skill set, 26 tools
Switch tools without switching habits — the methodology follows you.
Claude CodeCLI
CursorIDE
WindsurfIDECodex CLICLIGemini CLICLIKiroIDETraeIDEQoderIDEAiderCLIOpenCodeCLIQwen CodeCLIAntigravityIDEDeerFlow 2.0AgentVS Code · CopilotIDECopilot CLICLI
Hermes AgentCLIClaw CodeCLIOpenClawCLICodeBuddyIDECodeArtsIDEClineIDEKilo CodeIDECrushCLIZCodeIDEDeepSeek HarnessCLIReasonixCLIFAQ
Is superpowers-zh free?
Completely free. MIT-licensed open source, forever, with no paywall or subscription.
Which AI coding tools are supported?
26 tools: Claude Code, Cursor, Windsurf, Codex CLI, Gemini CLI, Kiro, Trae, Qoder, CodeBuddy (Tencent), CodeArts (Huawei), Aider, OpenCode, Qwen Code, Antigravity, DeerFlow, VS Code (Copilot), Copilot CLI, Hermes Agent, Claw Code, OpenClaw, Cline, Kilo Code, Crush, ZCode (Zhipu), DeepSeek Harness, Reasonix.
What makes superpowers-zh unique?
A fully localized, battle-tested methodology framework for Chinese developers: brainstorming, planning, TDD, debugging, and code-review skills, plus 4 China-native skills (code review / Git workflow / docs / commit conventions), adapted for 26 AI coding tools. MIT-licensed and free forever.
How does it take effect, and how do I check it works?
Restart your AI tool and open a new session. npx installs the skills and bootstrap files into your tool's directory (e.g. .claude/skills/); the tool loads the bootstrap at startup and then triggers the right skill at the right moment, with no manual invocation. To check: in a new session, say "Let's make a react todo list". If it is installed, it asks you questions to work out what you want before writing any code.
Can I install once for all projects? (global install)
Yes. npx superpowers-zh --global installs into the tool's user-level directory (e.g. ~/.claude/skills), shared across all projects; you only re-install once to update. Project-level takes precedence, global is the fallback — they coexist. Tools with global support (15 in total, all verified load paths per each tool's docs): Claude Code / Codex CLI / Qoder / Windsurf / Qwen Code / OpenClaw / OpenCode / Hermes Agent / CodeBuddy / CodeArts / Crush / ZCode / DeepSeek Harness / Reasonix / TRAE CN; other tools (incl. Gemini / Antigravity, which have their own global methods) should be installed per-project or via their docs.
Will it slow my AI down or upload my code?
No. Skills are on-demand Markdown: zero runtime, no network, no code or data upload — everything stays local.
How do I update or uninstall?
Use the same channel you installed with, and restart the tool afterwards. npx (project): re-run npx superpowers-zh in the project to get the latest; npx superpowers-zh --uninstall to remove. npx (global): npx superpowers-zh --global to update; npx superpowers-zh --global --uninstall to remove. Claude Code plugin marketplace: claude plugin marketplace update superpowers-zh, then claude plugin update superpowers-zh@superpowers-zh; remove with claude plugin uninstall superpowers-zh@superpowers-zh. For a tool's own channel (Codex plugin, Gemini extension, and so on), follow that tool's guide under Tools below.
It is installed but nothing happens. What now?
Check three things: which channel you installed with; that it is current (for npx, just re-run it; for the plugin marketplace, check the version in claude plugin list); and that you restarted and opened a new session. Don't install it for Claude Code twice in one project (marketplace plus npx). If it still fails, ask the diagnosing-superpowers skill to analyze the session that went wrong; it writes up a report. When opening a GitHub issue, include: tool and version, install command, OS, what you typed, and what it actually did.
Pair it with the methodology for 2× efficiency
"AI Coding in Practice · The Three-Volume Methodology" — full tutorials for 10 AI coding tools plus real-world pitfalls. Online book + PDF, free forever.
Learn AI coding free · aiOlaOla ↗ Read free ↗Ready to make your AI actually ship?
One command installs 21 battle-tested skills into your tool. Free, open-source, zero-dependency.
$ npx superpowers-zh