---
title: 'Agent Skill'
description: 'Install the Yorker agent skill so Claude Code, Codex, Cursor, and other coding agents know how to write, deploy, and debug monitors.'
section: 'AI & Agents'
canonical_url: 'https://yorkermonitoring.com/docs/ai/skills'
---

# Agent Skill

The Yorker skill is a small set of Markdown instructions that teaches a coding agent how to work with Yorker: the `yorker.config.yaml` schema, the order to run commands in, what it must not do without asking, and how to investigate a failing monitor.

It follows the open [Agent Skills](https://agentskills.io) format, so it works with any agent that supports skills. The agent loads the short main file when a task involves monitoring and pulls in the detailed references only when it needs them, which keeps your context window free for your code.

## Install

From your project root:

```bash
npx @yorker/cli skills install
```

This writes the skill to `.claude/skills/yorker/`, where Claude Code looks for project skills. Commit the directory so everyone on the team, and every agent, gets the same instructions.

| Agent | Command | Installs to |
|---|---|---|
| Claude Code | `yorker skills install` | `.claude/skills/yorker/` |
| Codex and other agents that read `.agents/` | `yorker skills install --target agents` | `.agents/skills/yorker/` |
| Any other agent | `yorker skills install --dir <its skills folder>` | `<dir>/yorker/` |
| Every project on your machine (Claude Code) | `yorker skills install --global` | `~/.claude/skills/yorker/` |

### Flags

| Flag | Description |
|---|---|
| `--target <agent>` | `claude` (default) or `agents`. Picks the conventional skills directory. |
| `--dir <path>` | Install into a specific skills directory. Overrides `--target`. |
| `--global` | Install into your home directory instead of the current project. Cannot be combined with `--dir`. |
| `--force` | Replace an existing install. Without it, the command stops if the skill is already there so local edits are not lost. |

## What is inside

```
yorker/
├── SKILL.md                      # when to use Yorker, the edit-validate-test-diff-deploy loop, hard rules
└── references/
    ├── config.md                 # monitor types, alerts, SLOs, channels, secrets, location IDs
    ├── browser-scripts.md        # Playwright scripts Yorker can run, step markers
    └── investigate.md            # finding out what is failing and why
```

The rules in `SKILL.md` are the part that matters most. They tell the agent to:

- validate after every edit and never deploy a config that has not validated,
- show you the `yorker diff` plan and wait for your go-ahead before deploying,
- never pass `--prune` or `--force` unless you ask,
- stop on drift and let you choose which side wins,
- keep secrets in environment variables and never ask you to paste a key,
- read assertions off the real response instead of guessing field names.

## Use it

Once installed, you do not invoke the skill by name. Ask for the outcome and the agent picks it up:

> Add monitoring for the new `/api/payments` endpoint.

> Are any of our monitors failing? If so, work out why.

> Turn `tests/checkout.spec.ts` into a Yorker browser monitor that runs every 10 minutes from the US and Europe.

See [Agentic Workflows](/docs/ai/agentic-workflows) for what each of these produces.

## Update

The skill ships inside the CLI, so it is always the version that matches the commands you have installed. After upgrading the CLI, refresh the copy in your repo:

```bash
npm install -g @yorker/cli@latest
yorker skills install --force
```

The skill also points the agent at the live Markdown docs and tells it that the docs win if the two disagree, so a slightly stale copy degrades gracefully.

## Agents without skill support

Print the skill and paste it into whatever your agent reads on startup, such as `AGENTS.md` or a rules file:

```bash
yorker skills show >> AGENTS.md
```

Print a single reference the same way:

```bash
yorker skills show config
yorker skills show browser-scripts
yorker skills show investigate
```

With `--json`, `skills show` returns the content along with the list of available references, so an agent can discover and fetch them itself.

## Related

- [AI Agents & Coding Assistants](/docs/ai/overview): choosing between the skill, the CLI, and the REST API.
- [CLI reference](/docs/reference/cli): every command the skill uses.
- [Markdown & llms.txt](/docs/ai/markdown-access): the docs the skill links to.