Eyal's Claude Code skills
A set of Claude Code skills for planning, implementing, reviewing, and merging software.
Architecture
A human talks to one orchestrator. The orchestrator runs workers, and each worker uses the skills below. The inspector watches how they all work.
Inside a worker:
Install
Run these commands inside Claude Code:
/plugin marketplace add eyalmutzary/skills
/plugin install eyal-skills@eyalmutzary
/eyal-skills:how-to-setup-eyal-skills
The first command tells Claude where the plugin lives.
The second installs the skills.
The third configures the optional extras:
- Claude settings and the ELI5 output style
- A personalized
CLAUDE.md
- RTK, so shell output is filtered before it reaches the model
- Eyal's status line
- The inbox pane, dev-command permissions, a keep-awake hook, and a twice-weekly worktree sweep
- Matt Pocock's planning skills
Restart Claude Code after setup.
Plugin skills start with eyal-skills: to avoid conflicts with other plugins. For example:
/eyal-skills:how-to-use-eyal-skills
Use that skill whenever you are unsure what to do next. Tell it where you are, such as: “I have a spec but no tickets.”
This plugin is installed directly from GitHub. It is not yet listed in Anthropic's public plugin directory.
Workflow (high level)
| Phase | Your attention | Main skills |
|---|
| Plan | Full | Context gathering (/grill-with-docs, explorers, /wayfinder), design → /to-spec, /how-to-explain-plan, /to-tickets + work breakdown |
| Implement | AFK | /how-to-implement-stack (/how-to-implement per PR: implement + review loop, /how-to-write-code, /how-to-debug-e2e, open PRs) |
| Review & merge | Half | Review stack, /how-to-babysit-pr, /how-to-fix-pr-comments, then plan the next batch |
Many sessions at once: run /how-to-orchestrate in its own session. It starts workers, answers their routine questions, and brings you only what needs you.
Planning steps are a menu, not a checklist. Use more of them for large or complex work. See /eyal-skills:how-to-use-eyal-skills for guidance.
Matt's plugin supplies /grill-with-docs and /wayfinder. The setup skill installs it when needed. This repo keeps its own copies of Matt's /to-spec and /to-tickets, so agents can call them.
Skills in this repo
| Skill | What it does |
|---|
how-to-setup-eyal-skills | Configure ~/.claude (settings, CLAUDE.md, RTK, status line, inbox, hooks, worktree sweep, Matt's plugin) |
how-to-use-eyal-skills | “What should I do next?” router for the workflow |
how-to-explain | Explain a concept, bug, or system at your altitude |
how-to-triage-new-task | Decide: implement now, plan first, or ask |
how-to-plan | Run the whole Plan phase: estimate, gather context, spec, plan, tickets, docs PR |
how-to-explain-plan | Walk an implementation plan step by step before coding |
how-to-review-plan | Review an implementation plan before you build it |
how-to-write-code | Coding conventions (shape, naming, frontend notes) |
how-to-write-prompt | Guidelines for reused prompts: skills, system prompts, CLAUDE.md |
how-to-implement | One change to one reviewed, verified PR |
how-to-implement-stack | Autonomous batch: reviewed, verified PR stack from a work breakdown |
implement | Build a scoped spec or ticket test-first, then review it |
how-to-debug-e2e | Prove changes locally (Playwright, a mirrored backend) |
how-to-QA | Prove a PR really does what it should, with a QA report |
how-to-match-figma | Turn a Figma design into an exact spec, then list UI gaps after each build |
how-to-babysit-pr | Drive PRs through CI and merge |
how-to-orchestrate | Master session that runs many worker sessions and escalates only what needs you |
how-to-inspect | Watch how the master and workers use the skills, and report gaps |
how-to-fix-pr-comments | Triage review comments as claims, not orders |
|
Output styles
| Style | What it does |
|---|
ELI5 | Plain, short answers; what happened, did it work, what next |
Applied via setup (outputStyle in settings-public.json), not by the plugin alone.
