DSH Vision Toolkit
Native image Q&A, long-screenshot OCR, interface restoration, grounding, and pixel comparison.
Turn images into structured evidence for text-only models, with OCR, layout, and semantic understanding.
Modlens gives DeepSeek Harness a native image-reading tool. It converts screenshots and images into structured evidence that the model can quote, reason about, and use for interface work without requiring a vision-native model.
Commands appear only when they are present in maintainer documentation.
npx -y @deepseek-ai/dsh plugin --profile web add @liustack/modlens@3.21.1This command pins 3.21.1, but npm now publishes 3.26.5. The command is quoted from the maintainer and is never rewritten here, so check the repository before using the older pin.
Pinned npm command documented by the maintainer. pnpm 11 holds back releases younger than 24 hours, which is why the version is named rather than @latest. See the versioned verification fields for the current independent verdict.
Maintainer claims, external observations, and registry-run tests stay separate.
the exact artifact installed, registered, loaded, and satisfied its direct peer contracts under the requested DSH version
Evidence is tied to the current repository state and may change after DSH or plugin updates.
Where each registry statement comes from.
Observable capabilities and package signals, not a formal audit.