dsh-todo-enhancer

When working with an LLM, replies often contain content that demands a decision from you: pick one of several approaches, approve a pending task list, or select an option before execution. Conventional heuristic matching (treating every numbered or checkbox list as actionable) either over-triggers on descriptive prose or silently misses genuine decision points, fragmenting your workflow. dsh-todo-enhancer consolidates such content into a single interactive checkbox table above the composer—let you tick items, edit text inline, add or remove rows, select all or none, rename the list title, and append a multi-line note—then fire one click to send, fill the draft, or dismiss, and the panel vanishes instantly.

The plugin relies on a contract-based marker protocol. The Host half injects a system-prompt rule instructing the LLM to wrap any decision-relevant content in paired HTML-comment markers; the Client half parses only marked blocks, so ordinary lists never trigger the enhanced UI, eliminating false positives by design. Rendering is strictly gated on session completion (session.running must be false); the panel stays hidden during streaming output. Reasoning blocks and tool-call blocks are never considered, keeping the source clean.

Built for DeepSeek Harness users who regularly ask the LLM to enumerate options or generate approval task lists, and who want zero false triggers without sacrificing reading clarity. Purely front-end rendering with a single service call, no persistence overhead, and all styles and registrations are reclaimed automatically upon uninstall.