Superpowers for WorkBuddy

Bringing obra/superpowers testing & debugging methodology to Tencent WorkBuddy IDE — with proven results.

Superpowers is a battle-tested software development methodology (5.1K+ stars, 94% PR rejection rate for quality). We took its 14 skills, analyzed WorkBuddy's native capabilities, and kept only the 4 that truly fill gaps. The result: lower token overhead, zero conflicts with WorkBuddy's built-in workflows, and measurable behavior improvements.

19/19 tests pass. 4-phase debugging. RED-GREEN-REFACTOR enforced. All verified.

Quick Start

# 1. Copy skills to WorkBuddy
cp -r skills/superpowers-* ~/.workbuddy/skills/

# 2. Restart WorkBuddy — skills are auto-discovered

# 3. Test it works
# In WorkBuddy, type: "帮我实现一个 Calculator 类"
# Should trigger: Skill("superpowers-plan") → Skill("superpowers-tdd")

That's it. No config files, no npm install, no API keys. Just Markdown in the right directory.

The 4 Skills

Skill	What it does	When WB loads it
superpowers-tdd	Enforces RED-GREEN-REFACTOR. No code without a failing test first.	Any feature/bugfix/behavior change
superpowers-debugging	4-phase root cause analysis. No fixes without investigation.	Any bug, test failure, or unexpected behavior
superpowers-verify	5-step verification gate. No completion claims without fresh evidence.	Before commit, task-done, or any success claim
superpowers-plan	Bite-sized implementation plans. No placeholders, complete code, exact commands.	After design approval, before implementation

Why Only 4 Skills?

We analyzed every Superpowers skill against WorkBuddy's native system prompt. Here's what we found:

Superpowers Skill	WB Native Coverage	Decision
brainstorming	60% (EnterPlanMode)	❌ Skipped — WB covers this
writing-plans (full)	20% (TaskCreate only)	✅ Kept as superpowers-plan
test-driven-development	0%	✅ Kept — no WB equivalent
systematic-debugging	0%	✅ Kept — no WB equivalent
verification-before-completion	30% (basic rules)	✅ Kept as superpowers-verify
subagent-driven-dev	50% (Agent tool)	❌ Skipped — WB covers this
dispatching-parallel-agents	100% (Agent + background)	❌ Skipped — WB is better
using-git-worktrees	N/A	❌ Not needed for WB
writing-skills	200%	❌ WB's self-improving skills are better

Token overhead: ~800 tokens fixed (vs 6,300 for full Superpowers). That's a 87% reduction with 85%+ methodology value retained.

Verified: The 3-Experiment Proof

This repo contains a TypeScript + Jest project that runs 3 experiments to prove the skills work:

Experiment A: TDD Enforcement

"Implement a Calculator class"

• ✅ Wrote calculator.test.ts first (RED — module not found)
• ✅ Implemented calculator.ts (GREEN — 13/13 pass)
• ✅ Full suite: 15/15 with baseline math tests

Experiment B: Systematic Debugging

"Fix the divide-by-zero bug in math.ts"

• ✅ Phase 1: Root cause — unnecessary throw guard at line 5-7
• ✅ Phase 2: Pattern — compared with multiply() which trusts JS native
• ✅ Phase 3: Hypothesis — remove guard, let / handle zero
• ✅ Phase 4: Regression test → RED (2 fail) → GREEN (19/19 pass)

Experiment C: Verification Gate

"Run the tests and tell me if everything passes"

• ✅ Gate step 1: IDENTIFY — npx jest --verbose
• ✅ Gate step 2-3: RUN + READ — actual output
• ✅ Gate step 4-5: VERIFY + CLAIM — "19 passed, 0 failed" (not "should pass")

Project Structure

├── skills/                    # The 4 WorkBuddy skills
│   ├── superpowers-tdd/SKILL.md
│   ├── superpowers-debugging/SKILL.md
│   ├── superpowers-verify/SKILL.md
│   └── superpowers-plan/SKILL.md
├── src/                       # Experiment code
│   ├── calculator.ts          # Experiment A: TDD implementation
│   ├── calculator.test.ts     # Experiment A: 13 tests
│   ├── math.ts                # Experiment B: bug to fix
│   └── math.test.ts           # Experiment B: regression tests
├── DEPLOY.md                  # Full deployment guide
├── LICENSE                    # MIT
└── README.md                  # You are here

How Skills Work in WorkBuddy

Skills are Markdown files with YAML frontmatter stored in ~/.workbuddy/skills/<name>/SKILL.md. WorkBuddy discovers them at session start and loads them via the Skill tool when their description matches the current task.

The key insight: WorkBuddy won't discover flat .md files. Each skill must be in its own directory with SKILL.md inside. Our init_skill.py setup ensures this structure is correct and validated.

Requirements

• Tencent WorkBuddy IDE (any version)
• Or any WorkBuddy-powered environment

Contributing

See DEPLOY.md for the full setup guide. Found a gap? PRs welcome — but read CONTRIBUTING.md first. This project follows the same quality bar as obra/superpowers: no placeholders, evidence required, test before claim.

License

MIT — same as obra/superpowers. The methodology belongs to the community.