dsh-screenshot-tool

Screenshot capture for DeepSeek Harness on Windows. Two entry points share one capture path:

• screenshot tool — the model calls it, gets the absolute PNG path plus pixel and byte size, and reads the image back with read_image.
• Camera button — a control in the composer tool row (next to the attach button). It captures the desktop, shows a preview panel, and offers download, copy-to-clipboard, and close.

Both go through a GET /screenshot route on the Harness web server that streams PNG bytes; the tool additionally writes a file into the session working directory.

Requirements

• Windows (uses System.Windows.Forms + System.Drawing through PowerShell).
• PowerShell: pwsh 7 is preferred, powershell.exe is the fallback. It looks for pwsh on PATH, then C:\Program Files\PowerShell\7\pwsh.exe, then Windows PowerShell.

Install

From the plugin market, or:

dsh plugin --profile <profile> add dsh-screenshot-tool

Restart the Harness once after installing, so the client module table is scanned again; the camera button appears on the next start.

Usage

Ask the model to take a screenshot, or call the tool directly:

{ "path": "evidence/login-page.png" }

path is optional. Relative paths resolve against the session working directory; the default is screenshots/shot-<timestamp>.png there.

The camera button captures straight away and never writes to disk unless you press Download.

What it touches

• Captures the whole virtual desktop (all monitors) via GDI.
• Spawns one PowerShell process per capture; the script is passed inline, and the output path through the DSH_SHOT_PATH environment variable.
• The tool writes the PNG to the path you choose; the button keeps the image in memory until download.

Limitations

• Windows only.
• Captures what is on screen, so occluded windows are not visible.
• Large screenshots are downscaled by read_image, which can make small text hard to read.