Superpower enhancements for BMAD Method and Claude Code.
Maintained fork of agilite-2025/superpowers — extended with BMAD v6.9 compatibility.
| Superpower | Description |
|---|---|
| Mental Model Execution | Trace 2-3 scenarios step-by-step before submitting code |
| Approval Workflow | No code changes without explicit user approval |
| Guardrails | No git, no Task/Agent spawning, brief responses |
| Quality Architect | Independent, adversarial code reviewer |
| SDET Role | Separates test design (QA) from test automation |
| Meta-Orchestrator | Fixes the system, not just symptoms |
- Developer - Mental Model Execution + strict guardrails
- Quality Architect - Blunt, evidence-based reviewer (NEW)
- QA Engineer - Testing + validation focus
- SDET - Test automation engineer (NEW)
- Orchestrator - Meta-role that improves role definitions
npx @tongsitat/superpowersAfter install (or when a new team member checks out the repo), confirm superpowers are active:
npx @tongsitat/superpowers verifyThis checks four layers:
- TOML override files exist in
_bmad/custom/ - SUPERPOWERS markers are present in each file
- BMAD's resolver confirms principles appear in merged agent output
- Scans for new
bmad-agent-*skills that aren't yet covered
Exits 0 if all good (CI-safe). Exits 1 with actionable output if anything is broken or missing.
The installer auto-detects your environment and picks the right mode:
BMAD v6 + Claude Code (detected by _bmad/ dir + .claude/skills/bmad-agent-*/SKILL.md):
🔍 Detecting environment...
✅ BMAD v6 (Claude Code) detected at: .claude/skills
📦 Injecting superpowers into BMAD v6 agents via _bmad/custom/...
✅ bmad-agent-dev (wrote bmad-agent-dev.toml)
✅ bmad-agent-architect (wrote bmad-agent-architect.toml)
⏭️ bmad-agent-pm not installed, skipping
✅ Added: .claude/skills/superpowers-quality-architect/
✅ Added: .claude/skills/superpowers-sdet/
✅ .claude/instructions.md (guardrails)
✅ .claude/roles/developer.md
...
🎉 Superpowers activated!
Writes non-destructive TOML override files to _bmad/custom/ — your existing BMAD config is preserved and BMAD updates will never overwrite superpowers.
Legacy BMAD (detected by src/bmm/agents/, _bmad/bmm/agents/, or .bmad/agents/):
🔍 Detecting environment...
✅ BMAD detected at: ./_bmad/bmm/agents
📦 Injecting superpowers into BMAD agents...
✅ dev.agent.yaml (Mental Model Execution)
✅ architect.agent.yaml (Quality Architect mode)
✅ sm.agent.yaml (Meta-Orchestrator)
✅ Added: quality-architect.agent.yaml (NEW)
✅ Added: sdet.agent.yaml (NEW)
✅ .claude/instructions.md (guardrails)
🎉 Superpowers activated!
Claude Code only (no BMAD):
🔍 Detecting environment...
✅ Claude Code detected
📦 Installing standalone superpowers...
✅ .claude/instructions.md
✅ .claude/roles/developer.md
✅ .claude/roles/quality-architect.md
✅ .claude/roles/qa-engineer.md
✅ .claude/roles/sdet.md
✅ .claude/roles/orchestrator.md
✅ .claude/project-context.md (template)
✅ CLAUDE.md
🎉 Superpowers activated!
Superpowers are injected via _bmad/custom/*.toml override files — BMAD's official customization layer. Your existing config is preserved and survives BMAD updates.
bmad-agent-devgains Mental Model Execution + approval workflow principlesbmad-agent-architectgains Quality Architect mode principles- New skills added:
superpowers-quality-architect,superpowers-sdet
Your existing agent YAML/MD files are patched directly:
- Developer now requires Mental Model Execution traces
- Architect has Quality Architect mode
- New agents:
quality-architect,sdet
Start a role session:
"Start Developer session"
"Start Quality Architect session"
"Start QA session"
Or just use Claude Code normally - guardrails apply automatically.
Before submitting code, trace 2-3 scenarios:
Scenario: User clicks "Submit"
Step 1: onClick → handleSubmit()
✅ Function exists
✅ Parameters match
Step 2: API call → POST /api/submit
⚠️ Missing error handling for timeout
Step 3: Response → update state
✅ State update correct
Agents MUST:
- Explain the problem
- List possible solutions
- Wait for explicit approval
- ONLY THEN write/edit code
Blunt, evidence-based reviews:
❌ BAD: "There might be a small concern..."
✅ GOOD: "The error handling is broken. Users see blank screens."
Every criticism includes:
- Problem summary
- File:line location
- Impact (Financial/User/Operational)
- Priority (CRITICAL/HIGH/MEDIUM/LOW)
Superpowers are marked with comments in every file they touch:
# --- SUPERPOWERS:START ---
...
# --- SUPERPOWERS:END ---
- BMAD v6: delete
_bmad/custom/bmad-agent-dev.toml(and similar) and remove.claude/skills/superpowers-*/ - Legacy BMAD / Standalone: delete the marked block from each patched file
- Node.js 18+
- BMAD v6+ (optional - works standalone with Claude Code)
MIT
Issues and PRs welcome at github.com/tongsitat/superpowers
This project is a maintained fork of agilite-2025/superpowers by @agilite-2025, extended with BMAD v6.9 compatibility and the verify subcommand.