Plugins

Adds a strip above the composer that lists sent-but-unread (steering) text messages and recalls them before the model reads them, one by one or all at once.

Reflective self-evolution plugin for local Qwen3.8 27B/3.8 Flash and online APIs: a coding-loop convergence steerer curbs Qwen3.8 27B/Flash repeated-verification deep-thinking to converge faster and cut thinking time; a five-state delivery guard detects and corrects stuck turns and infinite loops; model-agnostic host logic, mostly compatible with online APIs; per-use continual self-improvement with counter-reason self-checks, critic escalation, inverted-index memory retrieval, coding-loop convergence steering, and a five-state delivery guard.

Pauses the agent right before it sends the next model request of a tool round while keeping the composer usable: a bare Enter or the resume button lets it continue, and text typed meanwhile is sent along as a steering message; the pause never interrupts an in-flight API call and the model does not notice it.

A simple DSH queue panel for editing, removing, steering, reordering, and bulk removal.

Message-level full-text search across DSH sessions via an FTS5 trigram index (user / assistant / steering / tool with per-kind filters), deduplicated results across fork/continued sessions, read-only context preview with auto-scroll to the focused message, one-click jump to the real session, and incremental/full index rebuild with health check; zh/en i18n follows the Host locale service.

Telegram bridge for DeepSeek Harness: bidirectional sessions, conversation steering, home-channel routing, and TTS voice notes sent as Telegram audio messages.

Supervisor-style workflow orchestration over top-level sessions: confirmation-gated batch dispatch, team workstreams, steering running tasks, idle-wait collection and live model-route discovery across 11 task_* tools plus a /tasks slash command.

Native generation-time Mixture-of-Agents: every turn, reference models run the same session context in parallel as pre-run drafts and one aggregator executes with full tooling; config-as-model recipes, F mirror / R trimmed views, enforced subagent gating (dual-shape session indexing), derived per-recipe vision (vision refs relay image reads to text-only aggregators), tiered cadence with separate steering and patrol rosters plus frozen reuse between beats, degeneration guard dropping degenerate reference output, anonymized model face (failure notices carry no provider identity).

Adds an Assist button beside the composer send key. It scans the skills and MCP tools available in the session, ranks them against the text currently in the input box, and lets you inject the selected ones into the running conversation as a steering message.

Recovers a DeepSeek Harness turn that ended with reasoning only, with no text and no tool call, by steering the same turn one more step. Structure-only detection, so it costs no extra model call.

Feishu IM bridge for dsh: create a dedicated Feishu bot by scanning a QR code, get one Card JSON 2.0 notice per finished turn, answer tool approvals and structured questions from the chat, and drive a session from it — listing and switching the bound session with /sessions and /resume, steering a running turn, and choosing per-chat permission modes (read-only, workspace, Auto Review, full). Long output paginates instead of being cut. Owner binding: the QR scan binds the scanner; with manually entered credentials the first sender claims ownership and everyone else is refused.

Inject steering or correction into a running session while the model is thinking, or cancel the current turn, from a floating browser panel.

Keyboard shortcut that injects an encouraging message into the running dsh-TUI conversation: steer when a tool is in flight, interrupt-and-deliver when the model is only streaming text

Reorder the messages queued while a DSH agent is busy, from inside the official queue dock: a drag handle and per-row ↑ / ↓ buttons on the dock's own rows, applied on the host as one concurrency-checked inbox splice. The dock is decorated rather than replaced, so previews, inline editing, steering, removal and attachments stay as shipped. A move is refused when the queue has changed since the caller last saw it, when the row is not a user prompt (a goal round prompt, for example), or when the queue belongs to a subagent session.

Runtime governor for DeepSeek Harness: zero-progress steering, generation-bound verification evidence, evidence-aware completion guarding, and task-profile tool restriction.