Agent skills by Zhaoyang Jia: tooling for building, evaluating, and coordinating LLM agents, web page export, and Go workspace management. Each skill is a small folder with a SKILL.md. They work with Claude Code, Codex and other agents that read skills.
These skills used to live in separate repositories (
stitch-skill,runlens-skilland so on). They now live here; the old repositories point back to this one.
Pick one route. Installing both leaves you with every skill twice.
Claude Code
Register the marketplace once, then install the plugin:
/plugin marketplace add itamaker/skills
/plugin install itamaker-skills@itamaker
The plugin is a managed bundle that updates when this repository does.
Codex, and other agents
npx skills@latest add itamaker/skillsChoose the skills you want and the agents to install them on. To install one skill:
npx skills@latest add itamaker/skills --skill=<name>For tinkerers
The same npx skills installer writes the skills into your repo as ordinary files you own and can edit. Nothing updates behind your back; pull changes when you want them with npx skills@latest update.
| Skill | Needs |
|---|---|
| runlens, ragcheck, promptdeck, datasetlint | Their CLI: brew install itamaker/tap/<name>, or a release binary from the tool's own repository |
| webpage-to-pdf | Google Chrome and Python 3 (the converter ships inside the skill) |
| remote-browser | Node.js 22.18+ with TypeScript support (the skill installs one if yours lacks it), npm, curl and a Cloudflare account; the skill deploys its own Worker for you |
| stitch | Node.js (or Bun) and Stitch credentials: STITCH_API_KEY, or OAuth via STITCH_ACCESS_TOKEN plus GOOGLE_CLOUD_PROJECT. The SDK runner installs itself on first use |
| go-workspace-skill | Python 3, Git and Go, plus a workspace config file (the skill can create one) |
| shipyard | Python 3.11+ for its configuration and native setup helpers; Git and authenticated gh for merged-branch cleanup. Native roles are an optional, separate setup step. |
Every skill here is model-invoked: the agent reaches for it when the task matches, or you can call it by name.
Command-line tools and skills for building and evaluating LLM agents: reading traces, scoring retrieval, testing prompts and auditing datasets.
- runlens: Find failures, flaky tools and latency outliers in agent JSONL traces.
- ragcheck: Score retrieval runs with P/R/MRR/MAP/nDCG@k and check answers by word overlap.
- promptdeck: Render prompt templates, expand variant matrices and rank scored variants.
- datasetlint: Audit JSONL datasets for duplicates, train/eval leakage and label conflicts.
Work with web pages and web-based design tools.
- webpage-to-pdf: Save a live webpage as a PDF, pixel-exact or with searchable text.
- stitch: Generate and edit Google Stitch designs and turn them into app code.
- remote-browser: Cloud browser on your Cloudflare account for JavaScript pages and manual logins.
Everyday development workflow.
- shipyard: Coordinate Codex and Claude with scoped tests, model routing, CI acceptance, and safe branch cleanup.
- go-workspace-skill: Sync, build and test a multi-repo Go workspace from one config file.
Skills still being tried out live in skills/in-progress. They are not part of the plugin.