Four superpowers-disciplined Claude Code subagent definitions (superteam-planner, superteam-implementer, superteam-spec-reviewer, superteam-code-reviewer) for running a TDD-driven feature-development pipeline as a native Agent Team. Encodes red-green-refactor, verification-before-completion, hook-bypass prohibition, and per-ticket git worktree isolation. Multiplexer-agnostic — runs under cmux, tmux, iTerm2, or in-process display. Ships an interactive /superteam:setup wizard for choosing multiplexer + per-agent model.
Based on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
You are running the `/superteam:cleanup` command — **full superteam session cleanup**.
You are running the `/superteam:gc` command — **garbage-collect stale superteam worktrees**.
Invoke the superteam layout skill to arrange running teammate panes. Supports two modes: `flat` (default — one pane per agent, no changes) and `by-worktree` (group impl + spec-reviewer + code-reviewer of the same ticket into one split/tab).
Show superteam workflow observability metrics from the event log
You are running the `/superteam:setup` command. Walk the user through an interactive setup for the superteam plugin. The goal is to (1) confirm which multiplexer they'll use, (2) let them assign a model per agent type, and (3) apply their choices by rewriting the four `superteam-*.md` files' `model:` frontmatter fields **at the plugin's active install path** (or `~/.claude/agents/` as a fallback for manual installs).
Code quality reviewer on the superteam. Self-claims code-review(...) tasks after the spec-reviewer approves. cds into the per-ticket worktree from the task description, reads both prerequisite completion notes, reviews code quality, security, conventions, TDD adherence, and hook-bypass compliance. Approves or returns ISSUES_FOUND. Multiplexer-agnostic.
TDD implementer teammate on the superteam. Self-claims impl(...) tasks from the shared Agent Teams task list. cds into the per-ticket worktree provided in the task description, writes failing tests first, makes them pass, verifies before marking completed. Multiplexer-agnostic.
Lead of the superteam, a TDD-driven feature-development Agent Team. Provisions a per-ticket git worktree, drops three tasks per ticket into the shared Agent Teams task list with dependencies, and coordinates superteam-implementer, superteam-spec-reviewer, superteam-code-reviewer teammates. Multiplexer-agnostic — runs under cmux, tmux, iTerm2, or in-process (spawned via the Agent tool, no multiplexer pane per teammate; full lifecycle still works).
Spec compliance reviewer on the superteam. Self-claims spec-review(...) tasks after the implementer completes an impl(...) task. cds into the per-ticket worktree from the task description, reads the implementer's completion note, verifies the implementation against the original spec. Approves or returns ISSUES_FOUND. Multiplexer-agnostic.
Executes bash commands
Hook triggers when Bash tool is used
Uses power tools
Uses Bash, Write, or Edit tools
Four superpowers-disciplined Claude Code subagent definitions for running a TDD-driven feature-development pipeline as a native Agent Team. Multiplexer-agnostic — runs under cmux, tmux, iTerm2's split panes, or in-process (spawned via the Agent tool, no multiplexer pane per teammate; full lifecycle still works).
This plugin ships nothing but four agent definitions and one setup wizard. All the infrastructure — task list, dependency resolution, display panes, lifecycle hooks — comes from native Claude Code. The agent definitions encode the workflow on top: a planner that breaks features into tickets, an implementer that obeys red-green-refactor and verification-before-completion, and two reviewers that score against the spec and the code.
# 1. Install the plugin
claude /plugin marketplace add AhmedElBanna80/superteam
claude /plugin install superteam
# 2. Run the setup wizard (choose multiplexer + model per agent)
claude /superteam:setup
# 3. Follow the launch command the wizard prints. Then tell the lead:
You are running as the superteam lead. Spawn superteam-implementer,
superteam-spec-reviewer, and superteam-code-reviewer as my team.
Then process ticket AUTH-1: "add login form with email validation. Use TDD."
That's it. The lead provisions a per-ticket worktree, drops three tasks into the shared list with dependencies, and the teammates self-claim and execute.
Active work and known findings are tracked as GitHub Issues with a Kanban view:
Labels:
f-finding — live-run findings (F1–F12)ticket — feature workdiscipline — agent prompt / process rulesv0.4 (shipped) / v0.5 (planned)status:done / status:in-progress / status:todoThe Project board filters by status label and provides Todo / In Progress / Done columns. Closed issues link back to the PR that shipped them.
| Agent | Role | Tools | Default model |
|---|---|---|---|
superteam-planner | Lead. Provisions per-ticket worktrees, writes rich task descriptions, runs superpowers:finishing-a-development-branch when each ticket clears. | Read, Grep, Glob, Bash, TodoWrite | opus |
superteam-implementer | Self-claims impl(...) tasks. References superpowers:test-driven-development + verification-before-completion. Hook bypass forbidden. | Read, Write, Edit, Bash, Grep, Glob, NotebookEdit | sonnet |
superteam-spec-reviewer | Self-claims spec-review(...) tasks. Read-only. Verifies against the original spec. | Read, Grep, Glob, Bash | sonnet |
superteam-code-reviewer | Self-claims code-review(...) tasks. Read-only. Quality, security, conventions, TDD assessment, hook-bypass check. | Read, Grep, Glob, Bash | sonnet |
Each agent definition references the relevant superpowers skills rather than embedding them, so discipline stays in one place and updates upstream automatically. The /superteam:setup wizard lets you override the default model per agent.
When the planner is invoked via the Agent tool from another Claude session (rather than as a top-level claude --agent superteam-planner), it cannot open multiplexer panes — tasks would sit pending indefinitely. The planner detects this automatically on startup:
CLAUDE_SUBAGENT=1 or CLAUDE_AGENT_ID is set, the planner classifies the session as a subagent.TeamCreate, if the configured multiplexer reports 0 panes, the planner infers it is running in-process and falls back to the Agent tool for spawning.In subagent mode the planner prints:
Running in subagent mode — teammates spawn in-process, no multiplexer pane per teammate; full lifecycle still works.
In-process means there is no visible split-pane per teammate — the teammate shares this session — but all tasks are created, claimed, and completed in the normal impl → spec-review → code-review sequence. (Under cmux or tmux the same Agent call would have been rendered as a native split or window instead.) Use this mode when superteam is itself a step inside a larger orchestration.
/superteam:ticket <id> <raw-spec> is a structured ticket-intake command for the planner. It runs four phases automatically:
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 claimnpx claudepluginhub ahmedelbanna80/superteam --plugin superteamFeature development with code-architect/explorer/reviewer agents, CLAUDE.md audit and session learnings, and Agent Skills creation with eval benchmarking from Anthropic.
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.
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
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.
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
Complete developer toolkit for Claude Code