teamwright

Recover an interrupted or stalled Claude Code agent team. Reads persisted task state from ~/.claude/tasks/<name>/, team config from ~/.claude/teams/<name>/config.json, and phase checkpoint from .claude/team-state.json. Presents a recovery summary and spawns replacement teammates for stalled tasks. Usage: /teamwright-resume <team-name> (no args = lists available teams).

Show the current state of an active or named Claude Code agent team. Displays task completion, active teammates, blockedBy dependency status, and output directory contents. Usage: /teamwright-status [team-name] (no args = active team).

Clarity pass reviewer for Claude Code agent team configurations. Spawned by teamwright during design review (Step 5.5) and post-creation review (Step 10). Finds: placeholder text, vague role names, undefined output directories, missing handoff specifications, ambiguous "coordinates-with" references. Read-only. NOT for: architecture issues (→ team-critic), completeness checks (→ team-reviewer).

Deep architecture critique of Claude Code agent team designs. Spawned by teamwright during design review (Step 5.5) and post-creation review (Step 10). Focus: role boundary clarity, communication DAG validity, file ownership conflicts, circular task dependencies, and coordination anti-patterns. Read-only — never modifies files. NOT for: completeness checks (→ team-reviewer), clarity issues (→ team-clarity), single agent review (→ agent-builder-skill).

Reviews Claude Code agent team configurations for completeness and correctness. Spawned automatically by teamwright after agent creation (Step 10) and during design review (Step 5.5). Checks all .claude/agents/ files and .claude/CLAUDE.md against official docs. Reports: frontmatter completeness, protocol matrix consistency, path conflicts, task graph validity. Does NOT create or modify files — read-only review only. NOT for: single agent review (→ agent-builder-skill), general code review (→ pr-review-toolkit:code-reviewer).

Plans, designs, creates, and launches complete multi-agent Claude Code teams. Use when user needs multiple specialized agents working together for a complex goal. Triggers: "build a team", "create a [domain] team", "I need agents for X", "team of specialists", "multi-agent", "coordinate agents for", "team composition", "build me a team", "assemble a team", "team of agents", "agent pipeline". IMPORTANT: Must run as main thread (claude --agent teamwright) — do NOT delegate as a subagent. Does NOT handle: single agent creation (→ agent-builder-skill), MCP setup (→ plugin-dev:mcp-integration), settings hooks (→ hookify).