dsh-password-prompt

When an agent must run ssh, sudo, or any remote-login command, handling a password is awkward. Asking the user to switch back to a terminal breaks the workflow, while piping the credential into the model context means the secret lingers in reasoning traces and chat logs. dsh-password-prompt closes that gap with a masked HTML panel that pops up directly in the Web GUI. The agent calls the password_prompt tool, the browser shows a form with a password field (and optionally an account field), the user types and confirms, and the agent continues without any terminal interaction.

The security model is the standout feature. The password travels browser-to-host over RPC and is written to a private 0600 file on disk. The model only ever receives the file path, never the secret itself, so the credential cannot appear in reasoning or chat output. In account-plus-password mode the account is returned in the clear (treated as non-sensitive) while the password stays file-only. The agent then feeds that path into an askpass script for ssh or uses sudo -S with file redirection, and deletes the file immediately after. An optional companion skill nudges the agent to route every secret through this channel proactively, even before a Permission denied error surfaces.

It is aimed at developers and ops engineers who genuinely cannot replace a password with an SSH key but want the guarantee that the credential never enters the LLM context at all. The plugin ships as a bundle plus a dual-face client and activates on a stock DSH installation with a single command and no core patches.