In the DeepSeek Harness (DSH) plugin ecosystem, handling video data typically relies on external tools and lacks a localized, evidence-based end-to-end workflow. DeepDeck Video Sherlock is a local-first, evidence-based video investigation app plugin for DeepDeck.

This plugin is maintained by user jo32 and is designed to integrate video acquisition, transcription, and semantic frame indexing into the DeepDeck App Workspace. It is intended to be used when the DeepDeck App runtime services are mounted, and it relies on the app launcher, standalone app window, and scoped workspace provided by DeepDeck.

Core Features

The plugin implements video analysis by injecting Agent Skills. Its capabilities include:

  • Localized Processing: Local video acquisition, transcription, and semantic frame indexing. Large source videos, transcripts, indexes, and model outputs remain local by default unless explicitly published by the user.
  • Agent Skill Injection: Injects the analyze-video and video-sherlock-visualize Agent Skills to execute analysis tasks.
  • Native Playback Experience: Provides native video playback with HTTP byte stream support. The interface uses a responsive, monochrome design adapted to DeepDeck standalone windows.
  • Interactive Navigation: Supports a draggable timeline, chapter navigation, and clicking an evidence highlight to jump directly to the corresponding moment.
  • Context Synchronization: Synchronizes narrative context and displays recently inspected frames.
  • Data Visualization: Provides transcript density charts, topic indexes, an evidence wall, and limitation records.
  • Audit Reports: Generates complete Markdown audit reports, preserves original evidence, and lists explicit limitations.
  • Security and Isolation: Provides protected same-name artifact services through path containment checks.
  • Multilanguage Support: The UI supports Chinese and English and switches in real time according to DeepDeck language settings. Note: the UI language and investigation content language are controlled independently.

Installation and Enablement

There are two installation methods: installing through the DeepDeck interface (recommended) or performing a local development install.

Installing Through the DeepDeck Interface

  1. Open DeepDeck and go to Settings → Apps.
  2. Paste https://github.com/jo32/dsh-video-sherlock.git in Install an App plugin.
  3. Select Inspect source, confirm the detected package and build command, then click Confirm install.
  4. Restart DeepDeck as prompted.
  5. Open Apps → Video Sherlock in the sidebar.

Local Development Installation

Used for local development build and mounting the Cordis bundle:

git clone https://github.com/jo32/dsh-video-sherlock.git
cd dsh-video-sherlock
bun install --frozen-lockfile
bunx tsc --noEmit
bun run build
dsh plugin --profile web add "$PWD"

The final command only mounts the Cordis bundle; it is recommended to run it within a DeepDeck-managed web profile to ensure the required App runtime exists.

Typical Usage

  1. Analysis Workflow: Paste a public video URL or a local media path, optionally add investigation focus, and then start the full evidence-based analysis.
  2. Evidence Retrieval: Open a completed case, drag the video timeline or select an evidence highlight, and jump directly to the moment being inspected.
  3. Multidimensional Review: View transcript density signals, time-proportional narrative chapters, semantic topics, inspected key frames, and explicit limitations.
  4. Reporting and Archiving: Preserve authoritative transcripts, timestamps, and original artifacts while maintaining localized demos for specific scenarios.

Notes

  • Runtime Environment: This plugin runs with the permissions of the current DSH process. Review the source code and license before installing.
  • Dependency Version: Requires Bun 1.4 or later.
  • Data Storage: Generated investigations are stored under the video-analyses/ directory in the App Workspace and are intentionally excluded from Git.
  • Language Settings: The investigation content language is controlled by the “Content Language” field, while the UI language is controlled by DeepDeck language settings. The two are independent.

Summary

DeepDeck Video Sherlock provides DSH users with a closed-loop tool for completing video investigations within the DeepDeck ecosystem. It tightly integrates local processing and visualization with the DeepDeck runtime, making it suitable for scenarios that require handling sensitive video data or performing deep semantic analysis.