cloudcannon-cli

Original：🇺🇸 English

Translated

Use when running the CloudCannon CLI — logging in, listing or creating sites, reading and writing files on a hosted site, triggering and inspecting builds, changing a site's build configuration, reading form submissions, or finding out which command and flags exist for a task.

7installs

Sourcecloudcannon/agent-skills

Added on2026-10-30

NPX Install

npx skill4agent add cloudcannon/agent-skills cloudcannon-cli

SKILL.md Content

View Translation Comparison →

The CloudCannon CLI

@cloudcannon/cli

has two halves. The local half reads and writes files in the working directory and needs no account. The remote half talks to the CloudCannon API, needs credentials, and acts on live sites.

When to use

Finding out whether a command exists for a task, and what its flags are
Authenticating, or setting the CLI up in CI
Listing, inspecting, or creating sites and organisations
Reading or changing files on a hosted site, and committing those changes
Triggering a rebuild, or reading build and sync logs
Changing an existing site's build configuration
Listing form submissions

When not to use

Deciding what belongs in
cloudcannon.config.yml
— that is
```
cloudcannon-configuration
```
. This skill runs
```
configure
```
and
```
validate
```
; that skill owns what their output should say.
Running the site locally in CloudCannon —
```
cloudcannon dev
```
is covered end to end by
```
cloudcannon-dev-server
```
, including the rebuild loop and the
```
/__api
```
surface.
Calling the CloudCannon API from code — that is
```
cloudcannon-sdk
```
. It also owns the surface no command reaches: DAMs, backups, sync history and pulls, site scans, screenshots, provider connections,
```
site.copy
```
,
```
site.delete
```
.

Authoring editable regions or snippets —

cloudcannon-visual-editing

and

cloudcannon-snippets

.

The rule that costs the most

MUST confirm with the user before any command that writes to a hosted site.

Why: the remote half is not a sandbox.

sites files commit

pushes to the site's real git repository,

sites rebuild

starts a real build, and

sites update-build-config

changes how every future build runs. None of them prompt, and none are undone by re-running the command.

The read-only remote commands — every

list

, every

get

, every

print-*

— are safe to run unprompted.

Local and remote

Half	Commands	Credentials	Acts on
Local	`configure detect-*` , `configure generate` , `validate`	None	Files in the working directory
Local	`dev`	None	A built output directory
Remote	`orgs` , `sites` , `builds` , `inboxes` , `login` / `logout`	Required	Live sites over the API

MUST NOT assume a command is local because it reads rather than writes.

sites files get

reads, but reads it over the API from a hosted site.

Requirements

Node 24 or newer (the current LTS). Install globally with

npm i -g @cloudcannon/cli

, or invoke with

npx @cloudcannon/cli <command>

.

Quick start

sh

# What commands exist, and what flags does one take
npx @cloudcannon/cli --help
npx @cloudcannon/cli sites files commit --help

# Local: generate and check a configuration
npx @cloudcannon/cli configure generate --auto --initial-site-settings
npx @cloudcannon/cli validate

# Remote: authenticate, then read
npx @cloudcannon/cli login
npx @cloudcannon/cli sites list

File	Covers
SKILL.md (this file)	Routing, the local/remote split, the live-site rule
commands.md	The full command surface, and querying `documentation.json` for exact flags
authentication.md	Credential methods, their precedence, where they are stored, CI setup
editing-sessions.md	`sites files` — the staged-changes model, committing, discarding
troubleshooting.md	Symptom → cause → fix

Common mistakes

Excuse	Reality
"I'll write out the flags for this command from memory."	Flags change between releases. Query `documentation.json` or run `--help` . See commands.md.
"Build settings can only be changed in the dashboard."	`sites update-build-config` changes them for an existing site. See commands.md § Changing build configuration.
"I uploaded the file, so the site has it."	An upload lands in an editing session, not the repository. It is not live until committed. See editing-sessions.md.
"The command printed nothing, so it failed."	Read commands print JSON to stdout and nothing else. An empty result is an empty array, not an error — check the exit code.
" `--site` needs the UUID."	It accepts a name, ID, UUID, or domain.