# Install visualize

> Give your agent one instruction. It checks for an existing copy, proposes the right host path, asks before writing, and verifies discovery.

## Ask your agent

```text
Fetch https://usevisualize.ai/agent-setup.md and follow the instructions.
```

The runbook covers Cursor, Codex, Claude Code, OpenCode, Hermes, and Pi. It asks the agent to inspect first and get approval before changing files.

## Choose your agent

The runbook identifies the active host from runtime evidence first. If filesystem hints match more than one host, it asks rather than guessing. Each branch uses that host's tested, commit-pinned mechanism, expected path, reload step, and discovery check.

### Cursor

- How the agent recognizes it: Confirm the active runtime is Cursor. A .cursor directory alone is only a hint and is not sufficient.
- Expected path: `.agents/skills/visualize or ~/.cursor/skills/visualize`
- Mechanism: The reviewed commit-pinned Agent Skills installer.
- Install after approval:

```sh
npx skills@1.5.9 add display-dev/visualize@bdbf79c2b603badd8ae3bbc0231436142505377f --skill visualize
```

- Reload: Reload the Cursor window or start a new agent conversation.
- Verify: Ask the active host to list or discover installed skills and confirm visualize appears from the expected path.

### Codex

- How the agent recognizes it: Confirm the active runtime is Codex. AGENTS.md is shared by other hosts and is not sufficient evidence.
- Expected path: `.agents/skills/visualize or ~/.codex/skills/visualize`
- Mechanism: The reviewed commit-pinned Agent Skills installer.
- Install after approval:

```sh
npx skills@1.5.9 add display-dev/visualize@bdbf79c2b603badd8ae3bbc0231436142505377f --skill visualize
```

- Reload: Start a new Codex conversation so skill discovery runs again.
- Verify: Ask the active host to list or discover installed skills and confirm visualize appears from the expected path.

### Claude Code

- How the agent recognizes it: Confirm the active runtime is Claude Code. A CLAUDE.md or .claude directory alone is only a hint.
- Expected path: `.claude/skills/visualize or ~/.claude/skills/visualize`
- Mechanism: The tested Claude Code path through the pinned Agent Skills installer.
- Install after approval:

```sh
npx skills@1.5.9 add display-dev/visualize@bdbf79c2b603badd8ae3bbc0231436142505377f --skill visualize
```

- Reload: Start a new Claude Code conversation so skill discovery runs again.
- Verify: Ask the active host to list or discover installed skills and confirm visualize appears from the expected path.

### OpenCode

- How the agent recognizes it: Confirm the active runtime is OpenCode. AGENTS.md is shared by other hosts and is not sufficient evidence.
- Expected path: `.agents/skills/visualize or ~/.config/opencode/skills/visualize`
- Mechanism: The reviewed commit-pinned Agent Skills installer.
- Install after approval:

```sh
npx skills@1.5.9 add display-dev/visualize@bdbf79c2b603badd8ae3bbc0231436142505377f --skill visualize
```

- Reload: Restart the OpenCode session.
- Verify: Ask the active host to list or discover installed skills and confirm visualize appears from the expected path.

### Hermes

- How the agent recognizes it: Confirm the active runtime is Hermes and inspect its well-known design-skill mounts.
- Expected path: `.hermes/skills/visualize or ~/.hermes/skills/visualize`
- Mechanism: The pinned Agent Skills installer detects hermes-agent and installs the reviewed bundle without a fixed checkout or recursive overlay.
- Install after approval:

```sh
npx skills@1.5.9 add display-dev/visualize@bdbf79c2b603badd8ae3bbc0231436142505377f --skill visualize
```

- Reload: Restart the Hermes agent session.
- Verify: Run hermes skills list and confirm visualize resolves from the approved project or global scope.

### Pi

- How the agent recognizes it: Confirm the active runtime is Pi and that the pi package command is available.
- Expected path: `skills/visualize/SKILL.md inside Pi's installed Git package`
- Mechanism: Pi's native Git-package installer with the reviewed commit as its ref.
- Install after approval:

```sh
pi install git:github.com/display-dev/visualize@bdbf79c2b603badd8ae3bbc0231436142505377f
```

- Reload: Run /reload or restart Pi.
- Verify: Run pi list, then ask Pi to list or discover skills and confirm visualize resolves from the installed display-dev/visualize Git package.

## Manual fallback

If you prefer to install it yourself, use the reviewed release.

### Manual installation command

```sh
npx skills@1.5.9 add display-dev/visualize@bdbf79c2b603badd8ae3bbc0231436142505377f --skill visualize
```

Reload the agent, then ask it to list installed skills. `visualize` should appear.

Setup ends there: it does not create brand files, generate a document, sign in, or publish.

## Try your first request

Once setup is verified, continue in a conversation that already contains useful findings and ask:

```text
Turn the findings in this conversation into a visual report.
```

That creates a local HTML file. Publishing is a separate, optional request after you review it.

If the conversation is empty, give the agent a topic or local source path first. You can also ask it to create a **clearly labelled sample** so you can learn the workflow without mistaking demonstration content for real findings.