Multi-agent review of the current diff (or a PR/MR number, branch, or file target) for correctness bugs plus reuse, simplification, efficiency, altitude, and conventions cleanups, at a selectable effort level. Low/medium surface fewer, high-confidence findings; high/xhigh/max widen coverage (language pitfalls, security, concurrency, wrapper correctness) and may include uncertain findings. Use when the user asks to review a diff, branch, MR, or PR, or to hunt for bugs before committing. Pass --comment to post findings as inline PR/MR comments, or --fix to apply the surviving findings to the working tree after the review.
Produce a complete, navigable project-state documentation set for any software project (mobile, web, desktop, backend, or monorepo) — architecture, data, flows, API, tests, security, state/sync, performance, observability, platform parity, config, and health. Use when the user wants a full-codebase audit, architecture documentation, an onboarding "helicopter view", a regenerable project snapshot to diff over time, or a grounded top-10 risks / action-items report. Orchestrates parallel "scout" agents in two waves plus a synthesis pass, emitting markdown plus self-contained interactive HTML per section.
Review the changed code for reuse, simplification, efficiency, and altitude cleanups, then apply the fixes. Quality only — it does not hunt for bugs; use /code-review for that.
Plan, prioritize, and sequence implementation work so every iteration ends with a working version the end-user can already benefit from. Applies a "progressive JPEG" principle — start with a thin core scaffold that already delivers real value, then add features in order of user-visible value. Bugs in already-shipped functionality come first; engineering for hypothetical future steps is rejected. Triggers on "what's next", "plan the next steps", "prioritize features", "order of implementation", "MVP plan", "roadmap", "what should ship first", or any other request that orders multi-step work.
Scaffold or update a project to match the standard CLAUDE.md structure — root CLAUDE.md, docs/ tree (architecture, source-map, testing trio, workflow, gotchas, deploy, api), docs/ephemeral/ working area, and .claude/ agent + skill stubs. Use when starting a new repo, adopting the standard layout in an existing repo, or repairing a partial scaffold. Skips existing files unless --force.
Connect to a remote tmux session over SSH using iTerm2's native tmux control mode (-CC), creating the session if missing and detaching any stale client. Works on macOS with iTerm2.
Configure native macOS notification banners for Claude Code events (Stop, PermissionRequest, Elicitation) via terminal-notifier hooks, for both local macOS and remote Linux/tmux -CC setups bridged through iTerm2 Triggers. Use when the user asks to enable macOS notifications for Claude Code, be pinged when long tasks finish, get alerts for permission or MCP input prompts, or set up the iTerm2 Trigger that catches @@CLAUDE_NOTIFY markers from a remote server.
Decompose an approved plan document into self-contained task files under <plan-dir>/tasks/backlog/. Each task captures title, scope, out-of-scope, validation, tests, docs, and a final checklist — complete enough that a fresh agent can pick it up cold and ship it. Use when a plan is finalized and needs task breakdown before execution starts. Not for creating plans, not for doing the implementation work itself.