deploying-laravel-cloud

Files (2)

• SKILL.md
• checklists.md · reference/

SKILL.md (first lines)

# Deploying with Laravel Cloud CLI

## Setup

```sh
composer global require laravel/cloud-cli
```

Authentication is already set up (see Accounts). Only run `cloud auth -n` when `jq '.api_tokens | length' ~/.config/cloud/config.json` prints 0 and the project has no `.env.cloud`.

## Accounts

The user has more than one Laravel Cloud account. Projects use the default account, whose token is saved in `~/.config/cloud/config.json`. A project on another account has a `.env.cloud` in its root setting `LARAVEL_CLOUD_TOKEN`, which the `cloud` command on PATH loads before running the real CLI. Worktrees use the file of their main checkout.

- Never read, print, or echo `.env.cloud`, `LARAVEL_CLOUD_TOKEN`, or the saved tokens. `cloud auth:token --list` prints tokens in full, so do not run it.
- Never run `cloud auth` or `cloud auth:token --add` without asking. A second saved token makes every project without a `.env.cloud` fail with "Multiple API tokens found".
- Moving a project to the other account means asking the user to create `.env.cloud` in the main clone with `LARAVEL_CLOUD_TOKEN=<token>` and `chmod 600`. Never write the token yourself.

## Commands

Commands follow a CRUD pattern: `resource:list`, `resource:get`, `resource:create`, `resource:update`, `resource:delete`.

Available resources: `application`, `environment`, `instance`, `database-cluster`, `database`, `cache`, `bucket`, `domain`, `websocket-cluster`, `background-process`, `command`, `deployment`.

Some resources have additional commands (e.g., `domain:verify`, `database:open`, `instance:sizes`, `cache:types`). Discover these via `cloud -h`.

Never hardcode command signatures. Always run `cloud <command> -h` to discover options at runtime.

## CLI Flags

Always add `-n` to every command — prevents the CLI from hanging.
Never use `-q` or `--silent` — they suppress all output.

Flag combos per operation:
- Read (`:list`, `:get`) → `--json -n`
- Create (`:create`) → `--json -n`
- Update (`:update`) → `--json -n --force`
- Delete (`:delete`) → `-n --force` (no `--json`)
- Environment variables → `-n --force`
- Deploy/ship → `-n` with all options passed explicitly (no `--json`)

## Deployment Workflow

Determine the task and follow the matching path:

First deploy? → `cloud ship -n` (discover options via `cloud ship -h`)

Existing app? →
```sh
cloud repo:config
cloud deploy {app_name} {environment} -n --open
cloud deploy:monitor -n
```

Environment variables? → `cloud environment:variables -n --force`

Provision infrastructure? → `cloud <resource>:create --json -n`

Custom domain? → `cloud domain:create --json -n` then `cloud domain:verify -n`

For multi-step operations, see [reference/checklists.md](reference/checklists.md).

Not sure what the user needs? → ask them before running anything.

## When a Command Fails

1. Read the error output
2. Check resource status with `:list --json -n` or `:get --json -n`
3. Auth error? → the token is probably expired or revoked. Stop and ask the user for a new one
4. Fix the issue, re-run the command
5. If the same error repeats after one fix, stop and ask the user

Always run `cloud deploy:monitor -n` after every deploy. If it fails, show the user what went wrong before attempting a fix.

## Subagent Delegation

Delegate high-output operations to subagents (using the Task tool) to keep the main context window small. Only the summary comes back — verbose output stays in the subagent's context.

Delegate these to a subagent:
- `cloud deploy:monitor -n` — deployment logs can be very long
- `cloud deployment:get --json -n` — full deployment details
- `cloud <resource>:list --json -n` — listing many resources produces large JSON
- Fetching docs from https://cloud.laravel.com/docs/llms.txt via `WebFetch`

Keep in the main context:
- Short commands like `:create`, `:delete`, `:update` — output is small
- `cloud deploy -n` — you need the deployment ID immediately
- Any command where you need the result for the next step right away

## Rules

Follow exact steps:
- Flag selection — always use the documented combos above
- Deploy sequence — deploy then monitor, never skip monitoring
- Destructive commands — always confirm with user first, show the command and wait for approval
- Error loop — diagnose, fix once, ask user if it fails again

Use your judgment:
- Instance sizes, regions, cluster types — ask the user if not specified
- Which resources to provision — based on what the user describes
- Order of provisioning — no strict sequence required
- How to present output — summarize, show raw, or extract fields based on context

## Remote Access

### Tinker (>= v0.2.0)

Run PHP code directly in a Cloud environment:

```sh
cloud tinker {environment} --code='Your PHP code here' --timeout=60 -n
```

- `--code` — PHP code to execute (required in non-interactive mode)
- `--timeout` — max seconds to wait for output (default: 60)

The code must explicitly output results using `echo`, `dump`, or similar — expressions alone produce no output.

Always pass `--code` and `-n` to avoid interactive prompts.

From mateusjunges/dotfiles, MIT.

More from mateusjunges/dotfiles