claude-t3-devkit

Wire the claude-t3-devkit into an existing repo — enable the plugin, then generate tailored CLAUDE.md orchestration + agents (no scaffolding)

Implement a feature/change via domain-routed sub-agents, review the diff on Slack, then hand the approved branch to /ship

Scaffold a fresh create-t3-turbo monorepo and wire it to the claude-t3-devkit (settings, CLAUDE.md orchestration, agents)

Open a PR, run a multi-lens review, and notify Slack — with approval gates before opening and merging

Author CLAUDE.md orchestration rules and .claude/agents specialists tailored to this repo

Read-only review lens for architecture and design — boundaries, coupling, abstractions, error handling, and testability of a diff. The senior-engineer perspective in the ship review stage.

Read-only review lens for standards compliance — naming, structure, import order, and idioms measured against the repo's own conventions and lint/format config. Use in the ship review stage.

Read-only GO/NO-GO gate for adding or upgrading a dependency. Use proactively whenever a change introduces or bumps a package, before it is installed.

Read-only review lens for design-foundations compliance — does a UI diff use the project's exact colour palette, type faces/scale, logo rules, icon style/size, and spacing/padding/size scale, measured against the project's design rules file (e.g. `.claude/rules/design.md`). Mechanical and token-precise (cites the rule and the repo token), distinct from the holistic `design-reviewer`. Use in the ship review stage for frontend changes.

Read-only design review lens — holistic visual judgment on a UI diff: visual hierarchy & rhythm, layout & responsive integrity, motion, print fidelity, and overall brand feel (bold vs generic/timid). Judges the rendered screenshots, not just the JSX. Exact design-token/palette/type/spacing compliance belongs to the `design-foundations-reviewer`, not here. Use in the ship review stage for frontend changes.

Author a Claude Code "ship" pipeline that opens a pull request, runs code and security review via sub-agents, and posts Slack notifications — built on top of the subagent-orchestration baseline. Use this skill whenever the user wants to set up, configure, or improve an automated PR / review / notify workflow for a Claude Code project: a /ship (or /open-pr) slash command, specialist agents like pr-author, a multi-lens review panel (factual, architecture, security, consistency, redundancy), and slack-notifier, approval gates before opening or merging, Slack updates on PR events, or wiring these together with hooks and the gh CLI. Trigger it even when the user only describes the goal ("get my changes reviewed and tell the team on Slack", "automate opening PRs with a review step", "set up a ship command") without naming the pieces. For the routing rules, invocation protocol, and dependency safety this depends on, see the subagent-orchestration skill.

Author a Claude Code "/code-todo" workflow that implements a feature or change through domain-routed sub-agents, gates on a human Slack review of the diff, then hands the approved branch to the /ship pipeline. Use this skill whenever the user wants to set up or improve an implement-review-handoff pipeline for a Claude Code project: reading a change request, routing it to the right domain, dispatching implementer sub-agents with a full brief, posting the diff to Slack for approval, and handing off to /ship for PR/review/merge. It builds on the subagent-orchestration baseline (domain boundaries, invocation protocol, dependency safety) and the ship-workflow skill (slack-notifier, gh/gate/state wiring) — see those skills for the shared mechanics. Trigger it even when the user only describes the goal ("route my feature to the right subagent and ship it", "implement, let me review on Slack, then open a PR") without naming the command.

Author Claude Code orchestration configuration — the CLAUDE.md sub-agent routing rules and .claude/agents/ specialist definitions that teach a project's main session how to delegate. Use this skill whenever the user wants to set up, configure, audit, or improve how their Claude Code project uses sub-agents: writing or fixing CLAUDE.md routing rules, deciding when work should run in parallel vs sequentially vs in the background, defining persistent specialist agents, restricting agent tools or permissions, cutting token cost by routing sub-agents to lighter models, or fixing over- or under-parallelization and vague sub-agent invocations. Trigger it even when the user only describes the symptom ("my Claude Code spawns agents that step on each other", "how do I make my main session delegate better", "set up agents for my repo") without naming CLAUDE.md or .claude/agents explicitly.

1 hook across 1 event