dsh-plugin-plain-slides

An agent finishes a task and prints a wall of text: file names, code fragments, nested parens, system paths. If you read code, you scan it and know the outcome. If you do not, you close the window and still do not know whether it worked. dsh-plugin-plain-slides is built for exactly that gap: it takes over the reply renderer and re-lays every response as a short slide deck, one page for the conclusion, a few pages for the process with one line per step, and a final page listing any question the agent is waiting on you to answer. The full original text is one click away, and elapsed time, operation count, and token usage sit in the card header. Nothing is deleted; the reading order simply changes.

By default the plugin costs nothing. The slides are laid out locally from the original agent wording, with no model call at all. Press the plain-language button and only that turn report text is sent to your selected model, never the working process. A successful rewrite is cached, so asking again even after a reload or restart shows the plain version instantly with zero additional calls. Failures retry at most three times with a wider output budget each attempt, and the button can be disabled entirely if you never want the rewrite.

The plugin also tidies what the conversation shows around the card. Tool rows, reasoning folds, injected context, compaction markers, and retries are hidden by default so you are not scrolling past the agent output to find the answer. Approvals, questions, plan reviews, and deliverables stay visible in their proper homes. Errors and truncation are always kept. Everything hidden remains one click away in the Trajectory view.

This plugin is for people who use DeepSeek Harness but do not want to read code line by line: product managers, designers, ops engineers, developers new to AI tooling. The agent writes whatever it writes; the plugin simply gives you a second way to read it.