Turn Claude Code into a coordinated AI engineering team — with the full Superpowers methodology built in.
ClaudeMaxPower is a GitHub template that transforms Claude from a solo assistant into a full
AI engineering team — with hooks, skills, persistent session state (via .estado.md),
and an integrated adaptation of the obra/superpowers methodology (brainstorm → spec → plan →
subagent-driven development with strict TDD → two-stage review → finish).
It works in two modes:
- New Project: Install ClaudeMaxPower, brainstorm a feature, assemble a team, and ship
- Existing Project: Add ClaudeMaxPower and it creates a team tailored to your pending work
Clone it, run /max-power, and watch Claude operate at maximum capability.
# 1. Clone the template
git clone https://github.com/michelbr84/ClaudeMaxPower
cd ClaudeMaxPower
# 2. Run setup
bash scripts/setup.sh
# 3. Edit .env with your tokens
nano .env
# 4. Open Claude Code in this directory
claude
# 5. Activate maximum capability
/max-power
# Or jump straight to your goal:
/superpowers:brainstorming "new feature" # requires: /plugin install superpowers@claude-plugins-official
/fix-issue --issue 1 --repo michelbr84/ClaudeMaxPower
/assemble-team --mode new-project --description "REST API for task management"A few things live in GitHub repository settings rather than in the codebase, so git clone
cannot copy them. Do these once per repo right after cloning the template:
The repo ships with a ready-made ruleset at
.github/rulesets/main.json that:
- Requires all 7 CI jobs to pass before merging to
main - Blocks direct pushes, force-pushes, branch deletion
- Requires linear history (squash or rebase merges only)
- Requires PRs even for the repo owner (so CI always gates merges)
- Allows the Admin role (
actor_id: 5) to bypass in emergencies — adjust if you want stricter, or delete thebypass_actorsarray entirely
To apply it:
- Go to Settings → Rules → Rulesets
- New ruleset → Import a ruleset
- Select
.github/rulesets/main.json - Review the populated form and click Create
If you ever rename a job in .github/workflows/ci.yml, update the matching
required_status_checks.context value in the ruleset and re-import — otherwise the renamed
job will block every merge as a "waiting" required check.
Skills like /fix-issue and /review-pr need a GitHub token at runtime; the
batch workflows additionally need an ANTHROPIC_API_KEY. Local use reads these from .env.
For CI / GitHub Actions runs, add them under Settings → Secrets and variables → Actions.
Two scripts, distinct jobs:
| Script | Purpose | When to run |
|---|---|---|
scripts/verify.sh |
Checks your local environment — tools installed, .env populated, hooks executable, GitHub CLI auth |
Right after bash scripts/setup.sh, or whenever a session feels broken |
scripts/verify-ci.sh |
Runs the same lint and structure checks CI runs, with the same pinned tool versions (shellcheck v0.10.0, actionlint v1.7.7) |
Before every push — green here means green in CI |
# Pre-push: mirror CI locally
bash scripts/verify-ci.shverify-ci.sh auto-installs shellcheck and actionlint on first run into
~/.cache/cmp-verify/bin/ (delete that directory to force a fresh download).
It exits non-zero if any required check fails, so it slots cleanly into
pre-push hooks or local CI loops.
ClaudeMaxPower/
├── CLAUDE.md ← Project-wide Claude instructions (layered)
├── .claude/
│ ├── settings.json ← Hook config + Agent Teams enabled
│ ├── hooks/ ← Automated guards, quality gates, session state
│ └── agents/ ← Specialized sub-agents with persistent memory
├── skills/ ← Reusable AI workflows (invoke with /skill-name)
├── workflows/ ← Batch automation scripts
├── scripts/ ← Setup, verify, hook self-test, skill validation
├── mcp/ ← MCP server configs (GitHub, Sentry)
├── examples/ ← Working demo projects
└── docs/ ← Detailed guides for every feature
| Feature | What It Does |
|---|---|
| Superpowers Pipeline | Brainstorm → spec → plan → subagent-driven dev → two-stage review → finish |
| One-Command Bootstrap | /max-power installs, configures, and routes you to the right skill |
| Agent Teams | Assemble coordinated teams of specialized agents with /assemble-team |
| Layered CLAUDE.md | Project-wide + subfolder-specific Claude instructions with @imports |
| Hooks | Auto-run tests after edits, block dangerous commands, save session state |
| Strict TDD | Iron-law TDD via /superpowers:test-driven-development (install the Superpowers plugin) |
| Systematic Debugging | 4-phase root-cause process via /superpowers:systematic-debugging — never patch a symptom |
| Git Worktree Isolation | Safe parallel development with /superpowers:using-git-worktrees and /superpowers:finishing-a-development-branch |
| Sub-Agents | Specialized agents (code reviewer, security auditor, doc writer, team coordinator) |
| Batch Workflows | Fix multiple issues, mass-refactor, Writer/Reviewer pattern with worktrees |
| MCP Integrations | Claude reads GitHub issues and Sentry errors directly |
| Example Projects | Real code with intentional bugs to practice skills on |
┌──────────────────────────────────────────────────────────────┐
│ Claude Code + ClaudeMaxPower │
│ │
│ ┌───────────────┐ ┌───────────────┐ ┌──────────────────┐ │
│ │ CLAUDE.md │ │ Hooks │ │ Skills │ │
│ │ (context) │ │ (guardrails) │ │ (workflows) │ │
│ └───────────────┘ └───────────────┘ └──────────────────┘ │
│ │
│ ┌───────────────┐ ┌───────────────┐ ┌──────────────────┐ │
│ │ Agent Teams │ │ Workflows │ │ MCP │ │
│ │ (coordinator │ │ (batch/ │ │ (GitHub/ │ │
│ │ + teammates) │ │ parallel) │ │ Sentry) │ │
│ └───────────────┘ └───────────────┘ └──────────────────┘ │
└──────────────────────────────────────────────────────────────┘
ClaudeMaxPower integrates the obra/superpowers (MIT)
methodology directly as inlined skills. You get the full pipeline without a submodule or
plugin dependency. See docs/superpowers-integration.md
for the merged pipeline, decision tables, and migration notes.
Four Iron Laws enforced by the skills:
- No production code without a failing test first (
/superpowers:test-driven-development) - No implementation without an approved spec (
/superpowers:brainstorminghard gate) - No fixes without root-cause investigation (
/superpowers:systematic-debugging) - No merging with failing tests (
/superpowers:finishing-a-development-branchverification)
Attribution: see ATTRIBUTION.md.
Invoke any skill with /skill-name [arguments] inside Claude Code.
Pipeline skills (Superpowers methodology):
Install once with /plugin install superpowers@claude-plugins-official. The methodology
skills then live under the /superpowers: namespace.
| Skill | Command | Description |
|---|---|---|
| Brainstorming | /superpowers:brainstorming "user-auth" |
Collaborative design → spec (hard gate) |
| Writing Plans | /superpowers:writing-plans <spec-file> |
Break spec into bite-sized tasks |
| Subagent Dev | /superpowers:subagent-driven-development <plan> |
Fresh subagent per task + two-stage review |
| Systematic Debugging | /superpowers:systematic-debugging "..." |
4-phase root-cause process |
| Finish Branch | /superpowers:finishing-a-development-branch |
Merge / PR / keep / discard + worktree cleanup |
| Using Worktrees | /superpowers:using-git-worktrees |
Safe isolated git worktree |
| TDD | /superpowers:test-driven-development |
Strict Red-Green-Refactor with iron law |
If you type one of the legacy unqualified names (/brainstorming, /tdd-loop, etc.) the
/superpowers-redirect skill catches it and points you at the canonical command.
ClaudeMaxPower native skills:
| Skill | Command | Description |
|---|---|---|
| Max Power | /max-power |
One-command activation — installs, configures, routes |
| Assemble Team | /assemble-team --mode new-project --description "..." |
Assemble an agent team (brainstorming gate enforced) |
| Fix Issue | /fix-issue --issue 1 --repo owner/repo |
Read issue → failing test → fix → PR |
| Review PR | /review-pr --pr 42 --repo owner/repo |
Structured review → post comment via gh |
| Refactor Module | /refactor-module --file src/foo.py --goal "..." |
Safe refactor with test baseline |
| Gen Commit Message | /gen-commit-message |
Read staged diff, propose a Conventional Commits message (the deterministic secret / debug / large-file / linter checks now run automatically via the pre-commit-check.sh hook) |
| Generate Docs | /generate-docs --dir src/ |
Auto-generate API docs from source |
| Superpowers Redirect | /superpowers-redirect |
Catches legacy /brainstorming-style names and points you at the canonical /superpowers:* |
Hooks fire automatically — no prompting needed.
| Hook | Trigger | What It Does |
|---|---|---|
session-start.sh |
Session opens | Shows git context, previous session state, loaded skills |
pre-tool-use.sh |
Before any Bash command | Blocks dangerous commands, logs all commands to audit.log |
post-tool-use.sh |
After any file edit | Auto-runs tests for the modified file |
stop.sh |
Session ends | Saves session summary to .estado.md |
Agents are invoked by Claude as sub-sessions with specialized roles.
| Agent | Memory | Role |
|---|---|---|
team-coordinator |
project | Orchestrates agent teams — spawns, coordinates, synthesizes |
code-reviewer |
project | Strict code review — correctness, security, tests |
security-auditor |
project | OWASP Top 10 scan, dependency audit, secret detection |
doc-writer |
user | Generates README, API docs, guides — adapts to your style |
# Fix multiple GitHub issues in sequence
./workflows/batch-fix.sh owner/repo 10 11 12
# Writer/Reviewer pattern with git worktrees
./workflows/parallel-review.sh --feature add-search --task "Add search_tasks() function"
# Refactor across all matching files
./workflows/mass-refactor.sh --pattern "get_user" --goal "rename to fetch_user"
# Generate dependency graph
./workflows/dependency-graph.sh --dir src/ --output docs/deps.svg- Getting Started — prerequisites, setup, first run
- Superpowers Integration — merged pipeline, decision tables
- Agent Teams Guide — assembling and coordinating agent teams
- Hooks Guide — how hooks work, how to customize them
- Skills Guide — using and writing skills
- Agents Guide — sub-agents and persistent memory
- MCP Integrations — GitHub + Sentry setup
- Batch Workflows — headless automation patterns
- Parallel Workflows — Writer/Reviewer with worktrees
- Troubleshooting — common issues and fixes
- 14 Advanced Techniques — the techniques that inspired this project
- ATTRIBUTION — MIT credits for adapted content
- Solo developers who want to get dramatically more out of Claude Code
- Teams building repeatable AI-assisted engineering processes
- Claude Code power users exploring every advanced feature
- AI workflow builders looking for patterns to adapt
MIT — see LICENSE
- Fork the repo
- Create a branch:
git checkout -b feat/your-feature git commit— thepre-commit-check.shhook automatically scans staged changes for secrets (blocking), debug statements, large files, and linter issues (warnings); use/gen-commit-messageif you want help drafting a Conventional Commits message- Open a PR — the
review-prskill will help review it
Issues and ideas welcome at GitHub Issues.