Vercel Agent Skills
vercel-labs/agent-skillsVercel's official 9-skill pack: React and Next.js performance rules, composition patterns, UI and writing review, React Native, deploy and cost audit.
The WordPress project's skill pack for coding agents: 19 skills covering blocks, block themes, plugins, REST, Abilities API, WP-CLI, Playground and performance.
with caveats
Detailed, well-organized WordPress procedures from the official organization, but the license is not declared in a standard form and the README calls this an AI-generated v1.
check first: relies on an outside CLI · installs packages → security details
This pack gives a coding agent written procedures for WordPress work so it stops falling back on pre-Gutenberg patterns. A wordpress-router skill runs first. It calls a triage script (detect_wp_project.mjs) to classify the repo as a plugin, classic theme, block theme, core checkout or full site, notes the tooling and tests present, and then sends the agent to the matching domain skill using a decision tree in its references folder.
The domain skills cover Gutenberg block development (block.json, attributes, deprecations), block themes and theme.json, patterns, plugin architecture and security, the REST API, the Interactivity API, the Abilities API (with separate audit and verify skills), WP-CLI and multisite operations, performance profiling, PHPStan setup, Playground and Blueprint JSON, @wordpress/env, the WordPress Design System, and the Plugin Directory guidelines. Each SKILL.md follows the same layout: when to use it, required inputs, procedure, verification and failure modes. Most skills have a references/ folder with deeper docs, and a few include small Node helper scripts such as list_blocks.mjs and perf_inspect.mjs. The README says GPT-5.2 Codex generated the skills from official documentation and that WordPress contributors then reviewed and edited them.
19 skills averaging about 6 KB each, most with several reference docs, a handful of Node helper scripts, plus 52 eval files and CI in the repo.
Counted from the repository file tree. How we score depth.
This is a collection of WordPress development skills, mostly documentation plus local Node scripts that inspect your repo and print JSON. Some workflows run or install external tools (WP-CLI, @wordpress/env, Playground CLI via npx ...@latest), and the optional installer can write skills into global agent folders in your home directory. Two of the listed scripts (project triage and WP-CLI inspect) were not available for review, and the repo's maintenance scripts call WordPress, GitHub and Anthropic APIs from CI.
wp binary (core, option, doctor, profile commands) against the WordPress install given by --path/--url.skills/wp-performance/scripts/perf_inspect.mjs:32npm -g install; perf_inspect also suggests wp package install for the doctor and profile commands.skills/wp-env/SKILL.md:42npx @wp-playground/cli@latest server --auto-mount.skills/wp-patterns/SKILL.mdThe repository (193 files) on Nov 1, 2026. Not a full review: we read 11 of 13 code files; it relies on code outside the repository: skills/wp-project-triage/scripts/detect_wp_project.mjs (listed code file; full text not provided); skills/wp-wpcli-and-ops/scripts/wpcli_inspect.mjs (listed code file; full text not provided); WP-CLI (`wp` binary) and its doctor/profile packages.
NVIDIA SkillSpector static scan: 84 pattern hits (MCP Rug Pull 30, Agent Snooping 16, Data Exfiltration 11, Excessive Agency 10, Prompt Injection 8, Tool Misuse 7). The HIGH 'hidden instructions' hits are WordPress block-grammar HTML comments (<!-- wp:... -->) in pattern references. The rm -rf hit is CI cleanup of a /tmp test folder, and 'Always respond' is the JSON-format system prompt in the maintenance script. '--allow-root' is a WP-CLI flag that is passed only if the user supplies it, and the skill-enumeration and 'npx skills' hits are documentation paths and install instructions.
A review, not a certification. Skills change after we check them — read the files before installing.
GitHub reports the license as NOASSERTION, meaning it could not match a standard license. Check the repo's license section before commercial reuse or redistribution.
Not really. These are domain procedures and references the agent consults during interactive WordPress work; they define no long-running loop or resumable state of their own.
Vercel's official 9-skill pack: React and Next.js performance rules, composition patterns, UI and writing review, React Native, deploy and cost audit.
Cloudflare's official 19-skill pack for building on Workers, Durable Objects, the Agents SDK, Cloudflare One and more, with an optional bundled MCP server.
67 role-based developer skills for languages, frameworks, DevOps and security, each with reference files, plus 9 Jira/Confluence workflow commands.
A marketplace of Claude Code plugins for web builders: Cloudflare apps, design assets, frontend, dev tools, writing, Shopify and WordPress.
It is a set of 19 agent skills published by the WordPress organization on GitHub. The skills teach coding agents current WordPress practice for blocks, block themes, plugins, REST and Abilities APIs, WP-CLI, performance, PHPStan, Playground and wp-env. A router skill classifies your repo first and points the agent to the right skill.
Run npx skills add WordPress/agent-skills --skill wp-plugin-development for one skill, or list several names after --skill. Use --list to see every skill. The installer lets you choose project scope (for example .claude/skills/ or .cursor/skills/, which you can commit) or global scope in your home directory.
The SKILL.md compatibility fields say WordPress 7.0+ and PHP 7.4.0+. The Abilities API skill notes that projects targeting WordPress below 6.9 may need the Abilities API plugin or package instead of relying on core.
Yes, partly. The README says GPT-5.2 Codex generated them from official Gutenberg and WordPress documentation, and WordPress contributors then reviewed and edited them and tested them with AI assistants. The maintainers describe this as v1 and expect community fixes.
We found no credential reads, obfuscated code or hidden network calls in the skills. The scanner flags are documentation patterns such as block markup comments and a /tmp cleanup step in CI. The few helper scripts are small Node utilities for detection and inspection.