Anthropic's Skill Creator and Google's Workspace CLI: New Tools That Make AI Agent Skills a First-Class Workflow

Two releases dropped recently that enterprise IT teams should pay attention to: Anthropic published an official skill creator in their skills repository, and Google shipped gws — a Rust-based CLI for all of Google Workspace that includes 100+ agent skills and a native Gemini CLI extension.

These aren’t toy demos. They’re the tooling layer that turns “AI agent” from a buzzword into something you can actually operate.

What Shipped

Anthropic: The Skill Creator for Claude Code

Anthropic’s skill-creator is a meta-skill — a skill that creates other skills. It lives in the official anthropics/skills repo and works inside Claude Code (and Claude.ai for paid plans).

The workflow it enforces:

1. Capture intent — what should the skill do, when should it trigger, what’s the output format
2. Interview and research — edge cases, dependencies, input/output formats
3. Write the SKILL.md — structured frontmatter + markdown instructions
4. Test with evals — generate test prompts, run them, evaluate qualitatively and quantitatively
5. Iterate — rewrite based on feedback, expand test set, repeat

This is significant because it formalizes what was previously tribal knowledge. A skill is just a folder:

my-skill/
├── SKILL.md          # Required: frontmatter + instructions
├── scripts/          # Optional: deterministic/repetitive tasks
├── references/       # Optional: docs loaded as needed
└── assets/           # Optional: templates, icons, fonts

The SKILL.md frontmatter is minimal:

---
name: my-skill-name
description: A clear description of what this skill does and when to use it
---

The skill creator also handles eval infrastructure — it saves test cases to evals/evals.json and includes tooling for quantitative benchmarking with variance analysis. This is the part most teams skip and then regret.

Installing skills in Claude Code is now a first-class operation:

# Register the Anthropic skills marketplace
/plugin marketplace add anthropics/skills

# Install skill sets
/plugin install document-skills@anthropic-agent-skills
/plugin install example-skills@anthropic-agent-skills

The repo also ships production skills that power Claude’s document capabilities (DOCX, PDF, PPTX, XLSX) as source-available references — useful for understanding how complex skills are structured.

Google: The Workspace CLI (gws) with 100+ Agent Skills

Google’s gws takes a different approach. Instead of a skill-creation tool, they shipped the entire operational surface: a CLI that dynamically builds its command tree from Google’s Discovery Service, plus 100+ pre-built agent skills organized into four tiers.

The CLI itself is written in Rust, outputs structured JSON, and auto-discovers new API endpoints as Google adds them. No static command list to maintain.

npm install -g @googleworkspace/cli

gws auth setup                    # One-time OAuth config
gws drive files list --params '{"pageSize": 10}'
gws sheets spreadsheets create --json '{"properties": {"title": "Q1 Budget"}}'

The skills are organized into layers:

Install skills with the standard npx skills command:

# Everything
npx skills add https://github.com/googleworkspace/cli

# Just what you need
npx skills add https://github.com/googleworkspace/cli/tree/main/skills/gws-drive
npx skills add https://github.com/googleworkspace/cli/tree/main/skills/gws-gmail

Gemini CLI integration is a one-liner:

gws auth setup
gemini extensions install https://github.com/googleworkspace/cli

This gives Gemini CLI direct access to all gws commands and skills, inheriting your existing auth credentials.

MCP server support means any MCP-compatible client (Claude Desktop, VS Code, etc.) can use the same tools:

{
  "mcpServers": {
    "gws": {
      "command": "gws",
      "args": ["mcp", "-s", "drive,gmail,calendar"]
    }
  }
}

The enterprise features are notable: encrypted credential storage (AES-256-GCM with OS keyring), service account support with domain-wide delegation, headless/CI export flow, multi-account management, and Model Armor integration for scanning API responses for prompt injection before they reach your agent.

# Sanitize responses through Google Cloud Model Armor
gws gmail users messages get --params '...' \
  --sanitize "projects/P/locations/L/templates/T"

Why This Matters for Enterprise IT

Skills are becoming the standard unit of AI agent capability. Both Anthropic and Google have converged on the same pattern: a markdown file with structured frontmatter that tells an AI agent what it can do and how to do it. Anthropic calls them skills. Google ships them as skills. The Agent Skills spec is formalizing this across vendors.

This matters for three reasons:

1. Portability. Skills written for Claude Code work in Claude.ai, the Claude API, and (increasingly) other agent runtimes. Google’s gws skills work in Gemini CLI, Claude via MCP, and any MCP-compatible client. You’re not locked into one agent.

2. Governance. Skills are files in a repo. You can review them, version them, test them, and gate them through your normal change management process. The skill creator’s eval framework gives you a way to regression-test agent behavior — something most teams are doing manually or not at all.

3. Operational scope just expanded. Google Workspace is the productivity backbone for a huge portion of enterprises. Having 100+ pre-built, structured skills for Drive, Gmail, Calendar, Sheets, Admin, Vault, and Chat — with proper auth, pagination, and safety controls — means your AI agents can operate across your entire Workspace footprint without custom integration work.

What To Do About It

If you’re running Claude Code:

• Install the skill creator: /plugin marketplace add anthropics/skills
• Use it to formalize any repeatable workflow your team does. The eval infrastructure alone is worth the setup time.
• Review the document skills (docx, pdf, pptx, xlsx) as reference implementations for complex skills.

If you’re a Google Workspace shop:

• Install gws: npm install -g @googleworkspace/cli
• Start with the helpers (gws-gmail-triage, gws-calendar-agenda, gws-sheets-read) — they solve the 80% case immediately.
• Look at the persona skills for role-based agent configurations. The persona-it-admin and persona-exec-assistant bundles are particularly well-structured.
• If you’re using Gemini CLI, gemini extensions install and you’re done.

Regardless of platform:

• Treat skills as code. Put them in version control. Run the evals. Gate changes through PR review.
• Start building your organization’s skill library now. The teams that have reusable, tested skills when agent adoption hits critical mass will move significantly faster than those starting from scratch.
• Watch the Agent Skills spec — cross-vendor skill portability is coming, and investing in the current format is a low-risk bet.

The era of manually prompting AI agents for every task is ending. Skills are how you ship repeatable, testable, governable agent behavior. Both Anthropic and Google just made it dramatically easier to get started.