dsh-planchart

When you use an agent for project planning, the output tends to live inside chat bubbles -- structured in the model head but flat and scrollable in yours. dsh-planchart adds a real display surface: one planchart_set call and the browser renders two views at once. A steps panel docks to the right with phase groupings, status marks, and per-step detail (finished steps collapse to titles). Beside it, a PlanChart tab holds an auto-laid-out architecture diagram that exports as SVG or PNG, ready to drop into a doc or slide deck.

Iteration feels natural because it is dialogue-driven. You do not re-serialize the whole chart; you simply say "mark step 3 done" or "add a cache layer," and the next turn replaces the previous figure wholesale. The chart rides on tool/result.meta, so it survives restarts, resumes, and forks rather than vanishing with a session. The visual language is deliberately restrained -- a single accent color for the core, dashed borders for external dependencies, hairline strokes everywhere else -- so a nine-box graph stays legible instead of becoming a color contest.

This plugin is for developers already working inside DeepSeek Harness who want planning output to be glanceable and exportable. If you regularly need to turn an agent architectural or process breakdown into a diagram your team can read, or you want the planning surface to update as naturally as a conversation, it turns the chat-to-diagram step into a zero-effort byproduct.