Plugins

A New Session switch that puts the conversation into a throwaway Git worktree under the project's worktree pool (`<project>/.dsh/worktrees` by default), plus a panel to inspect and discard the copies.

Evolver for DeepSeek Harness, by EvoMap: recalls a proven strategy behind each prompt, reports its reuse at turn end, and distills verified work outside the conversation.

A desktop pet for DeepSeek Harness: an always-on-top character and a roster panel that show what the watched conversations are doing, with seven animations, Workspace grouping, subagents drawn under their parent, and one click to open a session. Ships Windows, macOS (Intel and Apple silicon) and Linux window binaries in the package, so installing runs no install scripts and needs no build step.

Derivative of cheesehaqi/dsh-qq-onebot-bridge with the same package name, so the two cannot be installed together. Bridges a personal QQ account into DeepSeek Harness over OneBot 11: private and group messages become per-conversation agent sessions and replies are sent back to QQ. Ships a reverse WebSocket server, allowlists (empty = deny all), @ parsing, image and SILK voice handling, and a separate local console on 127.0.0.1:8880 with the decision trace, offline replay, event injection, a read-only preflight and host/Waylay start-stop controls. Needs a separate OneBot v11 implementation (a third-party unofficial QQ client): the default, Waylay, injects into the official Windows QQ.exe (Windows 10/11 x64 only) and is distributed as a separate release archive, not bundled in this repo. botQqAllowed and replayGuardEnabled are edited in DSH Settings > QQ助手, and the replay guard is armed automatically when the logged-in account changes.

Bridges the official IAR C-SPY MCP server into DeepSeek Harness: build .ewp projects, start simulator or hardware debug sessions, and read variables, registers, memory, the call stack and target output from the conversation. Windows only, and it needs an IAR Embedded Workbench installation. Setup needs Python 3.10+, Git and network access to GitHub and PyPI.

Send DSH event notifications to Gotify, a generic webhook, or ntfy when a task finishes, is interrupted, needs approval, or needs an answer. The message carries only the session title, workspace name and event reason, never the conversation body, and a failed send is logged without interrupting the session.

Per-workspace DSH conversation status in the Windows system tray, plus native DeepSeek Harness notifications when a conversation needs approval, a choice or free-text input, completes, or fails. Windows 10/11 and DSH Desktop 0.2.3-rc.1+ only. The tray menu and notifications work only after a one-time desktop bridge install (user-level, reversible): it repoints the DSH shortcut at a hidden launcher that injects a hook into DSH's Electron main process through a short-lived loopback Node inspector; without the bridge the plugin only writes state.json. Idle conversations stay out of the tray by default; nine live options are editable in a Tray follower settings page.

Connects Octo direct messages, groups, and topics to DeepSeek Harness Web, with separate conversations, channel settings, and reply progress cards. Needs an Octo bot token and Node.js 24 or 26, and ships switched off until you enable it; private chats are admitted by a local allowlist or Octo friendship, and groups and topics need an explicit @mention.

Per-turn git checkpoints of a workspace: revert one turn’s changes (or go back to the end of an earlier turn), previewed file by file, and forget that part of the conversation from the model context.

Tongzhou: named local Codex–DSH discussions and read-only conversation views, with a DSH bundle and optional one-click Codex skill setup. Initially supports macOS with DSH Desktop 0.2.2-rc.1.

Render any Markdown report or DSH session excerpt into shareable PNG images with no npm dependencies, driving the Chromium-family browser already on the machine and falling back to PDF, then HTML. Reports come out as a summary card plus a full verbatim long image; a report written for dsh-fact-check also gets the tier badge, the two rails and the per-claim tiering, all taken from the report itself, and a report following one of the other report skeletons it defines gets the section that skeleton designates as its conclusion lifted onto the summary card. Conversations come out as a message flow, exportable by turn range, event range or the most recent N turns.

Inline audio player for generated clips and songs: the play_audio tool renders a player in the conversation (promoted into the transcript when the host supports it, otherwise inside the tool-call fold). Serves only allow-listed directories, supports Range seeking and HEAD, 512 MB cap per file.

Attachment store manager: per-conversation inventory, impact-aware deletion, and relocation of ~/.dsh/attachments to another disk (relocation tested on Windows 11 only).

A conversation-view tab for turning one local reference video into a human-curated shot list. ffmpeg scdet finds shot boundaries, each shot gets one timecoded thumbnail, you untick what you do not need, and the export writes frames.csv (UTF-8 BOM), shots.json, numbered JPEG frames and a labelled contact sheet. Local-only, no upload, zero runtime dependencies, 39 node --test tests including a real-ffmpeg integration case. Requires ffmpeg 5.1+ (with ffprobe) installed separately.

Skill usage radar for DeepSeek Harness: records every skill hit from the skill tool and /name gestures into a local JSONL ledger, ranks skills by hits in the last 7 days, greys out zero-hit skills with an archive suggestion, and expands a row into time, session, trigger and prompt. Ships a conversation-view panel with frontmatter model-invocation toggle, skill creation, and a recoverable .trash archive/restore. 52 zero-dependency tests; loopback-only, no external network.

Measured /pua for DeepSeek Harness: every pressure message gets a receipt comparing the last N completed model steps with the next N — tool calls, verification-class steps, repeated-call peak, failed calls and text volume — with a three-way verdict (changed approach / merely restated / no change) and a daily ranking of which sentence moved the work. Ships a conversation-view panel, a /pua command with list/status/receipts/rank/off, append-only local JSONL receipts (no tool arguments, content or session ids), 38 node:test tests and a zero-dependency build. Reports observed differences, not causality. Inspired by MichengAI/dsh-pua (independent code); it registers the same /pua command, so it cannot be installed alongside dsh-pua.

Local-first quick capture: select text in a DSH conversation to save it as a quote with its source (model and session), or jot quick notes from the sidebar; records stay in a local JSON file, with optional sync to a self-hosted service.

Full-text search over DSH conversation history: decompresses the persisted per-session session.v4.jsonl.zstd logs, matches user/assistant message text, and opens a hit at the matching message (archived sessions are unarchived first). Also binds Ctrl+F.

Adds a Delete conversation action to the conversation menu that permanently removes the session log, projection cache, and sidebar entry; refuses while a turn, subagent, job, or reminder is running.

Context profiles for new sessions: pick which tools and which system prompt each new conversation receives. Deferred tools stay as name+summary placeholders, keeping the tool list and the prompt prefix cache stable.

Floats conversations that are currently running to the top of their workspace session list, and keeps them there after they finish.

Adds a switch to the left sidebar footer that toggles the assistant between a rigorous professional register and a relaxed, humorous one by registering a late bilingual system-prompt section; the mode is profile-wide, persists in DSH_HOME, and applies from the next request.

Provider and account hub for DeepSeek Harness: several API-key accounts per provider with one-click switching and failover when a call dies on quota or credential errors, a per-account usage and quota panel in the conversation header, and built-in Grok Build device-code OAuth.

Replaces the composer model picker with a model + thinking-effort picker: arrow keys left/right cycle the effort level and arrow keys up/down switch model; an Auto level scores each turn by how far the conversation has drifted from the main objective and by repeated failures (one extra small judge call per turn), picks the effort from that score, and labels every finished turn with the effort it sent.