Deutsch
Ages quickly Review by: 2027-03 .md Source

Claude Skills — A Netlify Deploy Check Example

Stop reinventing the wheel — recurring agent workflows belong in a skill.

2026-03-12 · Stefan Wendel · Claude Opus 4.6

A skill is a reusable slash command for Claude Code — a Markdown file with instructions that the agent executes on demand. Instead of typing the same prompt over and over, you define it once as a skill and invoke it with /skill-name.

Here’s the workflow we want to turn into a skill: After merging a PR to main, Netlify triggers a build automatically. The question is always the same: Is the deploy done yet? Until now, I’d open the Netlify dashboard, dig through shell history for a curl command, or hit CMD-R 5 to 10 times. All of it breaks the flow. And flow is the new gold, so: never break the flow ;-)

The solution has two parts: Store the API token for the Netlify deploy status check securely on your machine — and package the entire workflow as a Claude Code skill, so it’s available anytime with a single slash command.

Part 1: Secrets in the macOS Keychain

macOS has a built-in, encrypted secret store: the Keychain. No Homebrew package, no external tool — the security CLI ships with every Mac.

Store a token (one-time):

security add-generic-password -s "netlify-api-token" -a "1001lessons" -w "YOUR_TOKEN"

Retrieve a token (in scripts or prompts):

security find-generic-password -s "netlify-api-token" -a "1001lessons" -w

That’s it. The token sits encrypted in the Keychain, protected by your login password or Touch ID. No risk of accidentally committing it. No .env file that needs to be in .gitignore and still ends up in the wrong repo eventually.

This pattern works for any API token, SSH passphrase, or other secret you need in the shell.

Part 2: Claude Code Skills

Claude Code supports custom slash commands — called skills. A skill file is just a Markdown file in .claude/commands/ that contains a prompt. When I type /check-deploy, Claude executes that prompt.

Create the file: .claude/commands/check-deploy.md

Check the current Netlify deploy status for 1001lessons.io.

Read the Netlify API token from the macOS Keychain:
security find-generic-password -s "netlify-api-token" -a "1001lessons" -w

Call the Netlify Deploy API and report:
- State (building / ready / error)
- Timestamp
- Deploy duration

Run it: Just type /check-deploy in Claude Code. The agent reads the token from the Keychain, calls the API, and reports the status. No copying curl commands, no opening dashboards.

Why This Is a Good Pattern

The combination of Keychain and skill solves two problems at once:

Secrets stay secure. The token never appears in a prompt or a file that could end up in a repo. The agent reads it at runtime from the Keychain — exactly like a shell script would.

Workflows become reusable. Instead of explaining the same API call every time, I define the workflow once as a skill. The agent has the context — it knows which project, which site, which API. I type a slash command and get the result.

The principle applies to any recurring workflow: checking CI status, querying DNS records, fetching Lighthouse scores. If you’re typing the same prompt for the third time, it’s worth making a skill.

Step by Step

  1. Create a Netlify Personal Access Token: netlify.com → User Settings → Applications
  2. Store the token in Keychain: security add-generic-password -s "netlify-api-token" -a "ACCOUNT" -w "TOKEN"
  3. Find your Netlify Site ID: Site Settings → Site details
  4. Create the skill: .claude/commands/check-deploy.md with the prompt template
  5. Test it: Type /check-deploy in Claude Code

Goodie on Top: Automatic Deploy Notifications

It goes one step further. In my setup, I don’t even need to call /check-deploy myself — Claude reports the deploy status automatically when the context calls for it.

The trick: My CLAUDE.md includes an instruction telling the agent to check whether the deploy went through after a PR merge or push to main. Claude knows the skill, knows the context — and runs the check on its own. No slash command needed.

There’s nothing magical happening under the hood. Claude Code reads the CLAUDE.md at the start of each session and sees the registered skills. When the situation fits — a PR was just merged, posts were just pushed to the content repo — the agent runs the skill autonomously and reports back: “Deploy is ready, took 45 seconds.”

That’s the real advantage of skills over shell aliases or bookmarks: The agent can not only execute them on command, but also proactively — because it understands the context in which they’re useful.