How to Test Your Changes in the google/skills Repository: 8-Step Local Development Guide

Install your modified skill locally with npx skills add google/skills, then use agents-cli run and agents-cli eval run to verify behavior before deployment.

The google/skills repository contains Markdown-based "skills"—executable recipes consumed by the Agents CLI. Because skills are not compiled code, testing your changes requires installing them into the CLI's runtime environment and exercising the full agent lifecycle locally. This guide walks through the complete development loop using commands verified in the repository source.

Install and Configure Your Local Environment

Before testing any changes, you need the repository cloned and the skill installed into your local CLI registry.

Clone the repository and create a feature branch:

git clone https://github.com/google/skills.git
cd skills
git checkout -b my-feature

Install the skill locally so the Agents CLI recognizes your modifications. As documented in README.md lines 11-15, the canonical command is:

npx skills add google/skills

This pulls the repository into the CLI's skill registry, making your updated Markdown files available for execution. After installation, verify the CLI recognizes the new version:

agents-cli info

Run the Agent Interactively

The primary way to test a skill is through agents-cli run, which starts the agent in a sandbox for interactive validation. The Build & Deploy Agents skill (skills/cloud/google-cloud-solution-build-deploy-agents/SKILL.md lines 233-235) explicitly recommends this command for local testing:

agents-cli run

This launches an interactive REPL where you can trigger your skill's logic and observe responses in real-time.

If you're testing a deployed endpoint rather than local logic, target it directly:

agents-cli run --url https://my-service-url

Evaluate Agent Output Automatically

Systematic evaluation ensures your changes meet quality standards. The CLI provides agents-cli eval run to execute the agent, capture execution traces, and grade outputs against reference models. The onboarding skill (skills/cloud/google-agents-cli-onboarding/SKILL.md lines 49-51) highlights this as step 5 ("Validate Quality") in the standard workflow:

agents-cli eval run

This command is particularly valuable for regression testing—run it before and after your changes to confirm behavior remains consistent or improves as expected.

Preview and Verify Deployments

Before pushing to cloud infrastructure, preview what would be deployed using the dry-run flag. The Build & Deploy Agents skill notes (skills/cloud/google-cloud-solution-build-deploy-agents/SKILL.md lines 231-233) that this shows deployment steps without executing them:

agents-cli deploy --dry-run

# or shorthand:

agents-cli deploy -n

Once deployed, verify the running service. The Guided GKE AI Migration skill (skills/cloud/google-cloud-solution-guided-gke-ai-migration/SKILL.md lines 226-229) demonstrates a concrete test pattern using port-forwarding and curl:

kubectl port-forward svc/my-agent-svc 8000:8000 &

curl -X POST http://localhost:8000/v1/chat/completions \
  -d '{"messages":[{"role":"user","content":"Hello"}]}'

Complete Development Loop Example

Here's the full iteration cycle for modifying and testing a skill:

# 1. Clone and branch

git clone https://github.com/google/skills.git
cd skills
git checkout -b feature/fix-auth-skill

# 2. Edit your skill file

vim skills/cloud/google-cloud-recipe-auth/SKILL.md

# 3. Re-install after each change

npx skills add google/skills

# 4. Quick interactive test

agents-cli run

# 5. Automated evaluation

agents-cli eval run

# 6. Preview deployment

agents-cli deploy --dry-run

# 7. If deployed, live test

kubectl port-forward svc/my-agent-svc 8000:8000 &
curl http://localhost:8000/v1/chat/completions -d '{"messages":[...]}'

Repeat steps 2-5 until the agent behaves as intended.

Key Files in the Testing Workflow

Path	Purpose
README.md	Installation instructions (npx skills add)
skills/cloud/google-agents-cli-onboarding/SKILL.md	Full CLI workflow documentation
skills/cloud/google-cloud-solution-build-deploy-agents/SKILL.md	run, eval, and deploy command references
skills/cloud/google-cloud-solution-guided-gke-ai-migration/SKILL.md	Deployment verification patterns
skills/cloud/google-cloud-recipe-auth/SKILL.md	Example skill for modification

Summary

• Install locally with npx skills add google/skills to make your modified Markdown available to the CLI
• Interactive testing uses agents-cli run for hands-on validation
• Automated evaluation runs via agents-cli eval run for systematic quality checks
• Deployment preview with agents-cli deploy --dry-run catches configuration issues early
• Live verification combines kubectl port-forward and curl for deployed service testing

Frequently Asked Questions

What is the Agents CLI in google/skills?

The Agents CLI is the command-line runtime that executes Markdown-based skills from the google/skills repository. It handles agent scaffolds, builds, evaluations, and deployments. Skills themselves contain no compiled code—the CLI interprets their logic and orchestrates execution.

Why do I need to run npx skills add after every edit?

The CLI maintains its own skill registry separate from your git working directory. The npx skills add command copies or links your repository into that registry, making your current files available to agents-cli run. Without reinstalling, the CLI continues using the previous version of your skill.

Can I test google/skills without deploying to the cloud?

Yes. The agents-cli run and agents-cli eval run commands execute skills in a local sandbox without cloud resources. Only the optional deployment preview (--dry-run) and live service testing steps require infrastructure—but even then, dry-run shows what would happen without making changes.

How do I know if my skill is installed correctly?

Run agents-cli info after installation. This displays recognized skills and their versions, confirming your modified files are registered. If your changes don't appear, repeat npx skills add google/skills from the repository root.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}