dsh-minimal-vision

In the pursuit of a streamlined AI workflow, the minimalist mode offers a stable and clean starting point, yet it often feels limited by its lack of vision capabilities. Traditional vision plugins frequently require registering new models or modifying the system prompt, inadvertently polluting the carefully maintained context. The dsh-minimal-vision plugin is designed to resolve this very conflict. It architects vision as an "on-demand service," intervening only when you truly need it via a simple Bash call, thereby perfectly preserving the purity and predictability inherent to minimalist mode.

Its core strength lies in its non-intrusive, intelligent integration. The plugin leaves your familiar toolset untouched, still exposing only persistent-bash and str_replace_editor. When you drag in an image or document (like PNG, PDF, PPT), it works behind the scenes to pass the local file path to the Agent via a collapsible context block. The Agent then makes a judgment call, like a human would: deciding autonomously to invoke the vision model only when visual evidence is required. The entire process is akin to adding a "magnifying glass" to a clean canvas—one that can be hidden at will. It provides powerful capability while maintaining the canvas's pristine state.

This plugin is especially suited for developers, data analysts, or researchers who are deeply invested in the DSH minimalist mode. If you frequently process visual information from charts, screenshots, or reports, yet place a high premium on workflow stability and context cleanliness, dsh-minimal-vision offers the ideal compromise. It allows you to safely expand your Agent's perceptual boundary without introducing complexity or pollution, letting you focus on core tasks rather than managing the tools themselves.