> ## Documentation Index
> Fetch the complete documentation index at: https://docs.devhelm.io/llms.txt
> Use this file to discover all available pages before exploring further.

# AI Skills

> Agent skills that let Cursor, Claude Code, and Claude Desktop configure monitors, investigate incidents, run status pages, and administer your workspace from chat

DevHelm ships a bundle of **agent skills** that give Cursor, Claude Code, Claude Desktop, and other Anthropic-skill-compatible hosts first-class knowledge of the platform. After installing them, an AI agent can answer "why is my checkout monitor red?" with a real forensic trace, publish an incident update to your status page, or stand up a new notification policy — all from the chat window, in your repo context.

Skills are bundled with the DevHelm CLI (`devhelm`) and installed with one command.

## What's in the bundle

| Skill                 | What it does                                                                                                                                                                                            |
| --------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `devhelm-configure`   | Create & manage monitors, alert channels, notification policies, resource groups, dependencies, secrets, webhooks, environments. Auto-detects Monitoring as Code (YAML or Terraform) vs imperative CLI. |
| `devhelm-investigate` | Read-only diagnostic: list check results, pull incident forensic traces, correlate failures across regions, query uptime. Never mutates state.                                                          |
| `devhelm-communicate` | Status pages: create pages, add components & groups, configure custom domains, manage subscribers, publish incident updates. Requires explicit human confirmation before any public-facing write.       |
| `devhelm-manage`      | Workspace administration: API keys (list, create, rotate, revoke), environments, entitlements/plan inspection, team roster.                                                                             |

Each skill is self-contained and follows Anthropic's progressive-disclosure convention — the top-level `SKILL.md` stays under 500 lines; resource-specific detail lives in `references/` files that the agent loads on demand.

## 60-second install

<Steps>
  <Step title="Grab your API key">
    Copy an API key from **[Settings → API Keys](https://app.devhelm.io/settings/api-keys)** — a starter key named `Default` is created for you during onboarding; you can create additional ones or rotate them anytime. All API keys start with `dh_live_` and have full access to the organization.
  </Step>

  <Step title="Install the CLI">
    ```bash theme={null}
    npm install -g devhelm
    # or run one-off: npx devhelm skills install
    ```
  </Step>

  <Step title="Install skills for your agent">
    ```bash theme={null}
    npx devhelm skills install
    ```

    This auto-detects Cursor, Claude Code, and other MCP-compatible hosts installed on your machine and copies the skill bundle into the right place. To target a specific host, pass `--target cursor | claude-code | codex | windsurf | gemini | generic | all`.

    For Claude Desktop, which ships skills as uploaded bundles:

    ```bash theme={null}
    npx devhelm skills bundle --out ~/Downloads/devhelm-skills.zip
    ```

    Then import the zip from Claude Desktop → Settings → Skills → Add skill.
  </Step>

  <Step title="Authenticate">
    Export your key so the CLI — invoked by the skill — can call the API:

    ```bash theme={null}
    export DEVHELM_API_TOKEN=dh_live_xxxxxxxx
    ```

    Or, if you prefer a local context file:

    ```bash theme={null}
    devhelm auth login
    ```
  </Step>

  <Step title="Verify">
    Open a chat in your agent and ask:

    > "Use the devhelm-investigate skill to list my monitors and their status."

    You should see the agent read `@references/check-results.md`, run `devhelm monitors list`, and return a structured summary. If it can't find the skill, run `devhelm skills doctor` to diagnose.
  </Step>
</Steps>

## Skills vs MCP server — which to use?

DevHelm offers two AI integrations. They're complementary, not exclusive.

|           | AI Skills                                                        | [MCP Server](/sdk/mcp/overview)                                                      |
| --------- | ---------------------------------------------------------------- | ------------------------------------------------------------------------------------ |
| Delivery  | Markdown files bundled with the CLI                              | Long-running server process                                                          |
| Protocol  | Agent reads `SKILL.md` and calls the `devhelm` CLI               | Agent calls tools over the [Model Context Protocol](https://modelcontextprotocol.io) |
| Best for  | Imperative workflows, Monitoring-as-Code edits, PRs, code review | Read-heavy dashboards, live context during a coding session                          |
| Transport | Local process (agent shells out)                                 | Local process (stdio) or hosted (`mcp.devhelm.io`)                                   |
| State     | Stateless                                                        | Stateless per request                                                                |

Run both together if you like. The skills know about the MCP server and will defer to it for pure read paths when it's installed.

## Safety rails baked in

Every skill enforces the same guarantees:

* **Never print API key values** into chat output, commits, or logs.
* **Never publish** a status-page incident update without explicit, single-turn confirmation.
* **Never edit `devhelm.yml`** without showing the diff first.
* **Always prefer `plan` / dry-run** before any `deploy` or apply.

See the [skill source in the skill repo](https://github.com/devhelmhq/skill) for the full list.

## Next

<CardGroup cols={2}>
  <Card title="Configure each host" icon="gear" href="/sdk/ai-skills/configuration">
    Cursor, Claude Code, Claude Desktop — the exact install paths, env vars, and verification steps.
  </Card>

  <Card title="Example prompts" icon="sparkles" href="/sdk/ai-skills/examples">
    Copy-paste prompts that hit each of the four skills end-to-end.
  </Card>
</CardGroup>