Harvey-Will/dsh-vision-analysis

Adds an analyze_image tool to DeepSeek Harness with eight analysis modes - describe, OCR, UI review, chart data extraction, object detection, comparison, code generation and debug - working against any OpenAI- or Anthropic-compatible vision endpoint or its bundled free one.

Adds an analyze_image(image?, images?, mode?, prompt?) tool that takes an absolute local path, an http(s) URL or a data:image/...;base64, URL — plus up to maxImages (default 2, max 4) for multi-image calls — and returns the vision model's text into the conversation. Eight modes ship with tuned instruction templates and per-mode defaults: describe (default), ocr, ui-review, chart-data, object-detect, compare, code-gen and debug, where ocr and chart-data return machine-readable JSON (lines, rows) and debug reports endpoint connectivity with the API key fully masked. Configuration covers apiFormat (OpenAI chat/completions or Anthropic messages), baseURL, apiKey (falling back to UNIVERSAL_VISION_API_KEY), model, fallbackModels tried on HTTP 429, defaultMode, maxImages, maxBytes, timeoutMs, maxTokens, temperature and per-mode overrides, all editable live from Settings → plugin configuration. The README also documents result caching and retry with exponential backoff, and states that image bytes never enter the session log or reach the main model.

Vision & Multimodal ★ 1
View on GitHub ↗

Install

dsh plugin --profile web add github:Harvey-Will/dsh-vision-analysis

Copied verbatim from the README's "Quick start" block, whose first line is commented "From GitHub (no npm needed)". The npm registry check returned no package entry for this repo, so the GitHub form is what ships; the README clearly requires restarting the web profile afterwards.

Compatibility

Declared for DSH 0.1.0-rc.x through 0.2.0-rc.x (verified on 0.2.0-rc.1) with Node ^22.19 || >=24, and the quick start says to restart the web profile after installing. The analyze_image tool works with zero configuration against a bundled free anonymous vision endpoint (OVHcloud AI Endpoints, Qwen2.5-VL-72B). The image bridge, by contrast, needs two setup steps: add the model to bridgeModels in the plugin config and declare image in that model's inputModalities in settings.yaml.

Details

Alternatives

lijian-ui/dsh-vision-toggle · TZHR-invest/dsh-plugins#dsh-vision-tool · ethanweave/dsh-vision-better

More plugins in Vision & Multimodal

Browse more in Vision & Multimodal

Guides for Vision & Multimodal plugins