The design philosophy of DeepSeek Harness (DSH) is “everything is a plugin”. When building agents or handling day-to-day workflows, screenshots and clipboard images are often scattered across different software and require manual archiving. This plugin monitors the clipboard, pops up a system-level floating window near the mouse cursor, supports instant OCR and nightly AI organization, and turns screenshots directly into Obsidian note entries.

Core Features

  • Windows system-level floating window: After detecting a new screenshot (or copied image), it shows a topmost small window near the mouse.
  • Annotation and key marking: The popup includes an annotation input box and a “Key” checkbox. When “Key” is checked, that annotation is written to the note under a # **重点** heading.
  • Save Document / Save Image: Supports saving screenshots to the Obsidian vault and generating a merged note for the day; or saving only the image (with a #图片 tag).
  • Instant OCR: Uses the host LLM for recognition by default (zero configuration); if the host is unavailable, it automatically falls back to the Qwen API.
  • Self-healing assistant: If the resident clipboard assistant command exits unexpectedly, it automatically restarts with a backoff strategy (2s→30s, up to 5 attempts) and raises an alert.
  • Same-screen deduplication: The Windows screenshot tool occasionally writes the same screen to the clipboard twice; the plugin compares grayscale fingerprints and pops up only once within a specified window (default 6 seconds).
  • Nightly AI organization: For the day’s inbox entries, it triggers AI organization via a command, including categorization, generating category notes, bidirectional links, a daily summary, and archiving.
  • Web GUI configuration interface: DSH Settings → “Screenshot Ingestion”; changes take effect in real time.

Installation and Enabling

The plugin is published as a DSH plugin. The installation command is as follows:

dsh plugin --profile desktop add github:wangzhanchao883/dsh-screenshot-capture

It takes effect after restarting DSH. Clipboard monitoring and the floating window are automatically started by the plugin host (DSH must be running).

Configuration

On first use, configure the Obsidian vault path in the Web UI; otherwise, the capture function is automatically disabled.

DSH Web UI → sidebar Settings → Screenshot Ingestion section, configure directly:
- General: enable monitoring, Obsidian vault path, polling interval, cooldown time.
- OCR: recognition channel (host host LLM, default; or off disabled), host model (leave empty to follow DSH’s current default model).
- Floating window: horizontal/vertical offset, maximum preview width.

Note: The panel no longer includes Qwen configuration options. The Qwen key is read from the DASHSCOPE_API_KEY environment variable or the configuration file, and is used only as a fallback channel when the host model is unavailable.

Method 2: Configuration File

Configuration file location: %USERPROFILE%\.dsh-screenshot-capture\config.json

{
  "enabled": true,
  "vaultPath": "D:\\path\\to\\obsidian-vault",
  "ocr": { "mode": "host", "hostModel": "" },
  "pollIntervalMs": 200,
  "cooldownMs": 2000,
  "dupWindowMs": 6000
}

Usage Steps

  1. Keep DSH running.
  2. Use Win+Shift+S to take a region screenshot (or Ctrl+C to copy any image).
  3. The floating window appears near the mouse: enter content in the annotation box (optional), check “Key” if needed, then choose an action [Copy Screenshot / Save Document / Save Image].
  4. In the evening, tell DSH “Organize today’s screenshots” and select which entries to keep as prompted.

Environment and Considerations

  • Dependencies: Requires Windows PowerShell 5.1 + WinForms.
  • DSH version: Requires DSH 0.1.7+.
  • OCR conditions: When using the host LLM for OCR, DSH must load the llm and attachments services, and the current model must declare inputModalities: [ text, image ].
  • Capture disabled: If vaultPath is not configured, the capture function is automatically disabled and an alert is raised; no directories are created.
  • Data security: Clipboard images are processed locally only and are sent to the recognition channel only when recognition is performed. The plugin contains no built-in keys.

Summary

This plugin provides DSH users with a complete closed-loop workflow from screenshots to structured notes, leveraging the host LLM for zero-configuration OCR and supporting nightly batch organization. The source code and documentation are hosted on GitHub and are licensed under the MIT License.