dsh-desktop-tools

Desktop Tools for DeepSeek Harness

简体中文 · DeepSeek Harness · Security

Native Windows desktop automation for DSH agents, using UI Automation (UIA) where possible and scoped screenshot coordinates when applications do not expose semantic controls.

Features

• Five tools for window discovery, accessibility/control snapshots, screenshots, clicking, and typing.
• Per-session/window references and screenshot references prevent stale-coordinate reuse.
• Honors the current DSH session permission preset, plugin write toggle, per-process policy, and native approvals.
• Optional visible operation indicators; supports a PowerShell-based helper instead of a continuously running service.
• Automatically prunes eligible plugin-owned screenshots after the configured retention period, without deleting audit logs or active-session artifacts.

Requirements and install

Windows only. Requires PowerShell 7, a compatible DSH host, and the dependencies described in package.json. UIA or keyboard interaction requires an unlocked interactive desktop.

dsh plugin --profile desktop add github:Kerberos255/dsh-desktop-tools

Use your actual DSH profile name if different. Pin a commit for reproducible installations and restart DSH after installing/updating plugin code.

Quick start

1. Open Settings → Plugins → Desktop Tools. Review the allowed-process policy and leave writing disabled until needed.
2. Call desktop_windows to find a window, then desktop_snapshot to obtain current control references.
3. Use desktop_click or desktop_type with a current reference. Refresh the snapshot after changes.
4. For custom-drawn apps, use desktop_screenshot and its screenshotRef; coordinates are inside that screenshot, not global screen coordinates.

Permissions and privacy

Read-only sessions cannot write. Full-access DSH sessions may avoid repeated approvals for ordinary permitted interaction, but explicitly forbidden processes, secure desktops, credential windows, and sensitive submissions remain protected. An approval setting that disables prompts is not a grant of permission.

Credential fields are suppressed in supported snapshots/screenshots. Audit files and screenshots are stored only in the local runtime directory. Screenshot cleanup defaults to seven days; active sessions and audit records are excluded. See SECURITY.md and policy.json.

Tests

Run npm test for portable security/policy checks. True UIA, DPI, focus, keyboard and on-screen marker tests require an unlocked Windows host and cannot be inferred from GitHub Actions alone. Settings template: config.example.json.

Related: Browser Tools for web pages.

MIT licensed. See LICENSE.

Optional plugin diagnostics

The plugin optionally emits privacy-restricted operation and failure metadata to dsh-plugin-diagnostics. Diagnostics contain event type, status, fixed error code and elapsed time; conversation content and credentials are not logged. The diagnostics plugin is optional and does not change the host requirements.