Secure Remote Execution

Prepare a server for an AI worker’s first SSH task.

Understand host identity, SSH keys, remote permissions and recovery before giving an AI worker a real WordPress maintenance job.

Prepare the account, server identity and first task before connecting. Use the setup instructions for your installed Outloop release to configure the connection.

Follow the visual Server Access setup guide →

Prepare an account and a staging target

Confirm that your hosting plan supports SSH and which tools it provides. Create a dedicated account with access to the intended staging installation. Check filesystem permissions, database authority and any sudo privileges. Record the site directory, expected URL, owner and allowed task without putting credentials in the brief.

Understand the two identities SSH checks

A server host key identifies the server. Obtain its fingerprint through a trusted hosting console or administrator and compare it during connection setup. A fingerprint change requires investigation before accepting the new identity.

The user’s public key is registered for the remote account. The private key proves possession and must stay protected. Outloop’s model keeps that key in host-side secure credential infrastructure. Keychain protects stored credentials on the Mac; it does not reduce the remote account’s permissions.

Assign the connection to the correct Workspace

Open Server Access, choose Connect a server, verify the server identity and use Test connection. Authorize the client under Workspaces that may use it. The connection screen shows readiness and the last test result.

Confirm the client, environment and connection assignment before allowing execution. Verify connection discovery and authorization in the intended Workspace, including refusal from a Workspace without access. Configure and validate another client independently.

Start with an inspection you can verify

On a staging installation with WP-CLI available, a first task can ask the worker to report the WordPress version and plugin inventory. Supply the approved installation directory and prohibit changes for this initial task.

wp core version
wp plugin list --fields=name,status,version --format=table

These are WP-CLI examples to run in the confirmed installation directory, not Outloop tool-call syntax. Check installed extensions and command behavior before use. Avoid printing configuration files or broad logs that may contain credentials or personal data.

Authorize a bounded fix

Choose a staging issue with a known verification step. Confirm a usable backup or Git revision, required database backup and recovery owner. Have the worker inspect the relevant files, explain the change, execute within the agreed scope, and verify the browser result alongside command output. Promote to production only under separately established authority.

Handle long-running work and interrupted connections

Record the execution identifier, command outcome and website checks available in the approved release. If a connection drops, inspect server state before retrying a command that may already have changed files or data. Confirm process survival, cancellation and status recovery with the Product team before relying on unattended long-running work.

Report what actually finished

A useful completion report identifies the client and environment, summarizes changes, records checks and calls out uncertainty. Treat command output as potentially sensitive. Approve any screenshots or logs before sharing them outside the client’s authorized team.

Questions operators ask

Can Claude Code or another worker use SSH?

Agents with an execution tool can work through SSH. Their credential placement, execution environment and permissions depend on the integration. Confirm Outloop’s supported runtime and installed release before using the broker route.

Can the worker upload and download files?

Yes, within the remote account’s permissions. Confirm the source and destination, check the transferred file, and inspect the result before retrying a failed transfer.

Does protecting the SSH key protect every server secret?

The remote account may be able to read secrets already on the server. Use server-side permissions and appropriately scoped accounts to control that reach.