Efficiency
Efficiency plugins reduce round-trip friction: edit system prompts, reroll a turn, watch tokens, tap a suggested reply. They usually don't add dangerous tools.
dsh-prompt-studio edits system prompt sections with live preview—safer than blind-editing prompt fields in cordis preset YAML: you see what the model actually ingests this turn. Writers and anyone locking "no shell" should clarify rules here before UI writing panels. There is no standalone repo named "headline" in the public catalog; APPSO's /headline example is a capability demo, not a plugin name. This site doesn't list demos as products.
dsh-message-edit adds branched message edit, reroll, retry, and version timeline. When the Agent drifts, you don't need a new session. It doesn't fight append-only session logs—the log remains; you edit the branch fed back to the model. dsh-live-stats shows TPS, latency, context fill, cache hits, and tokens near the composer—including from the dsh-web-ui bundle.
dsh-suggested-replies puts chips above the Web composer. dsh-paste-input / dsh-drag-and-drop / dsh-input-history handle paste, drag paths, and input history. Task boards live in dsh-web-ui—not a separate efficiency repo; use the UI bundle for kanban.
Efficiency plugins are mostly safe to enable globally. Low risk, but they may touch localStorage and branch pointers. For benchmarks aligned to minimal mode, disable suggested replies and prompt studio so extra text doesn't enter context. V4 Flash public Code Agent scores used minimal + Max; efficiency plugins break that alignment—by design, not bug.