Use the StoreFront CLI with OpenClaw
OpenClaw is an agent runtime that reads project skills from the workspace's .agents/skills/ folder, the same folder Codex uses. There is no openclaw target; use --target codex. Status: Works with caveats. OpenClaw has discovered the skill and used its deeper references in a Linux Docker container (node:24). No macOS or Windows run has been confirmed.
Prerequisites
- Node.js 24 or newer, with npm, where OpenClaw runs.
- OpenClaw installed, with model credentials set up.
openclaw onboardneeds an interactive terminal. Its Claude CLI option needs Claude Code installed and signed in where OpenClaw runs; an API key is the alternative. - An UltraCart account whose user has StoreFront access.
Install the CLI and the skill
npm install -g @ultracart/storefront-cli@beta
uc-storefront --version
uc-storefront init my-storefront --target codex --no-input
openclaw setup --workspace my-storefront
The workspace can be a plain folder. OpenClaw does not need a Git repository, a trust step, an allowlist or a restart to find the skill.
Confirm OpenClaw sees the skill
cd my-storefront
uc-storefront skills status
openclaw skills info build-ultracart-storefronts
openclaw skills info should report the skill as ready and visible to the model. OpenClaw's own documentation describes its project skill path.
Start the agent
Onboarding leaves the OpenClaw Gateway running, and openclaw agent --local is then refused. Run openclaw agent without --local, or stop the Gateway first.
Sign in
Where the sign-in goes depends on the machine OpenClaw runs on. On a machine with an operating system keychain, such as a Mac, a Windows PC or a Linux desktop with Secret Service, the CLI keeps the sign-in in that keychain and no credential key is needed: auth status shows a keychain store, and you can go straight to the sign-in commands below.
In a container or cloud environment there is no keychain, so the CLI keeps the sign-in in an encrypted file. Give the container the credential key and a persistent configuration folder as described in Docker, Fly.io and CI runners, then check it:
uc-storefront auth status
In a container you want credential store: encrypted-file (durable, available) and credential key: set. Then the agent signs in in two steps:
uc-storefront profile create my-store
uc-storefront auth login --profile my-store --no-wait
uc-storefront auth login --profile my-store --continue --wait
uc-storefront profile bind my-store
The first command prints a URL and code and exits at once; you approve in your own browser. The second waits up to 45 seconds and exits 3 while still waiting, so the agent runs it again until it exits 0.
A container's outgoing address can change when it is redeployed, and each new address needs your approval. See Approved addresses and a stable IP.
Give OpenClaw the key
OpenClaw's shell commands inherit the Gateway's environment, and OpenClaw can set variables for it. Prefer a key file over the key itself, so the key never sits in a configuration file:
- Create the key file yourself, on storage other than the configuration folder that holds the sign-in:
umask 077; openssl rand -base64 32 > /path/outside/the/volume/uc-storefront.key. - Set only its path,
ULTRACART_STOREFRONT_CREDENTIAL_KEY_FILE=/path/outside/the/volume/uc-storefront.key, in~/.openclaw/.env, or for this skill alone underskills.entries.build-ultracart-storefronts.envinopenclaw.json(skills config). - Start a new agent run, then check with
uc-storefront auth status.
These settings do not reach a sandboxed run. If OpenClaw runs commands in a sandbox container, mount the key file into it and set the path there. Avoid agents.defaults.sandbox.docker.env for the key itself: its values are visible in the container's Docker metadata.
Caveats
- The confirmed setup is Linux in a Docker container with one model. Other operating systems and models have not been confirmed.
--target codexis the target to use. Removing it removes the copy OpenClaw reads, unlessgemini(Antigravity or Gemini CLI) is also a target.
Next steps
- Connect your store: profiles, binding and how long a sign-in lasts
- Safety and human approval: read before the first live change
- What it can do: the areas OpenClaw can work on with the skill