Secret masking
Command output often contains secrets: a token in an environment dump, a password in a config line. WECOMMANDER masks them on the device, before the result leaves it, so the AI does not see them.
What is masked
- Private key blocks.
- Authorization header values and credentials inside URLs.
- Well-known token formats: GitHub, GitLab, Slack, AWS access key ids, sk- keys and JWTs.
- The values of NAME=value and "name": "value" pairs whose name mentions a password, secret, token, API key or credential. The name stays visible; the value becomes [REDACTED], and the result says how many values were masked.
Where
Output of process_exec, process_list, task_status, terminal_read and terminal_read_page on agents; log_query and docker_container_logs; command output and process lists on SSH servers (in your app, then again on the gateway); scheduled task output; and audit text.
What is not masked
- File reads. Editing tools need the exact content, and you chose which folders the AI may read: keep secret files outside the allowed folders.
- Secrets in a form the patterns do not recognise. Masking is best effort, not a guarantee.
Turning it off
On by default. On an agent run webrising-commender --config <path> capabilities --redact-output off and restart the service; in local mode start with --no-redact. policy_info reports whether it is on.
- Security model in depth: The layers that keep AI access under control: outbound-only agent, allowed folders checked on the device, signed single-use approvals, safety net, secret masking, encrypted sync, OAuth and browser sessions.
- Agent: install, update and options: The WECOMMANDER agent on Windows, macOS and Linux: how it is installed and updated, the user it runs as, allowed folders, read-only and no-command switches, --allow-root, and checking its status.