Plugins

Windows-only derivative of the official ego lite DSH plugin (@ego-lite/dsh-plugin, citrolabs/ego-lite) that cannot be installed alongside it: connects DeepSeek Harness to the installed Windows ego lite app through authenticated loopback HTTP for task spaces, navigation, page reading, clicks, form input and screenshots.

Turn the Edifier Halo PixelBar desktop pixel display into a DSH tool: push short text, switch scenes, set screen color, ambient light and volume over HID, and restore the clock on demand. USB-connected device only; tested on Windows only; needs Python 3.8+ with hidapi (pip install hidapi).

Disk, CPU/memory/GPU, listening-port and token-usage cards with a full-page zoom; the port card works on Windows only and the GPU card needs nvidia-smi.

Rewrites bash tool calls through `rtk rewrite` on the `tools/execute` hook, so it applies to any provider route. Zero dependencies and no build step: a single ESM module that adopts exit codes 0 and 3, runs the original command otherwise (no equivalent, deny rule, missing rtk, spawn failure), restores the original arguments after dispatch, and honors `RTK_PREFIX_ONLY` for direct `rtk <cmd>` prefixing and `DSH_RTK_BIN` to pin the binary.

Chinese accounting and tax knowledge base for DSH agents, served by the hosted Quaesto service at api.accountingllm.site (needs network access; no account or API key, but your queries leave the machine): answers questions on CAS enterprise accounting standards, the small-enterprise accounting standards, CSA auditing standards and tax law with original articles, decision procedures, applicability checks and tax-accounting differences, exposed through 10 MCP tools and a china-accounting-standards skill.

One-shot OMP subagent provider for DeepSeek Harness — each delegation spawns the local `omp --print --mode json` CLI running with the user's own ~/.omp configuration and model bindings, and delegations are serialized so a single-session local engine is never oversubscribed. Requires the OMP (oh-my-pi) CLI installed locally and DSH 0.2.2-rc.1.

Per-session control of which skills and MCP tools/servers the model can see (built-in tools are untouched): default shows everything, auto asks the model each turn what to keep, smart picks from your saved groups, or tick groups by hand, with a switcher under the composer and fail-open to full visibility on any error. The default state is auto, which adds one model call per turn (the session model unless judgeModel is set).

CNC (NC/CNC/G-code) toolpath recognition, 3D display and a large editing view with quote-to-composer for DeepSeek Harness.

Session install radar for DeepSeek Harness: lists what each conversation installed (pip, npm, git clones, winget-like installers) with size, path, creation time, purpose, why it was installed, and what removing it costs. Ships a sidebar panel, 10 agent tools, a timestamp sweep of temp directories, a process-command-line audit, and a two-step confirmed cleanup limited to temp directories. An optional scheduled cleanup of old temp directories is off by default.

Manually resets old AgentRouter Responses replay state for chats blocked by upstream resource-binding errors while preserving history; supports DSH 0.2.2-rc.1.

Automatically split long DSH conversations into volumes (every 10 user messages by default) with todo handoff snapshots, source navigation and read-only archives.

Session-specific sticky cards with optional, explicitly authorized agent follow-ups. Targets the DSH desktop/web profile (protocol 0.2.2-rc.1); installed from GitHub.

Per-workspace Fork library: a bookmark on each finished assistant turn stores that turn boundary as a full compressed snapshot that survives deletion of the source session; a "Fork 库" conversation tab lists a workspace's forks as a tree (a fork saved from a launched session nests under its parent) with name, turn count, last question and answer, note and source link, and "从这里启动新会话" seeds a new session from any fork, repeatable. Each fork then names itself from its own context: the snapshot is replayed to the model with a naming request appended, and that exchange is never written into the snapshot or a session.

Per-session supervision with role assignments and handoff workflows; off until enabled with /supervise. Requires @harzva/dsh-role-registry 0.1.0-alpha.1 (not on npm) installed first from its GitHub tag archive.

DSH SMOOTH: config-only bundle for a lighter DSH UI - compact transcript and usage views, 16 px font and 200 ms job polling out of the box - plus a troubleshooting SKILL.md and opt-in recipes (appended by hand to your own cordis.patch.yml or the profile) to remove the Agents-Anywhere "Reconnect" button, slim the tool list, drop legacy sub-agent bundles and make compaction thresholds take effect. User patches still win; rollback = remove the bundle.

Automatic context compaction: a background observer distills early history into summary blocks, and when usage crosses the trigger line the engine commits whole blocks until it is back under the target line, with no model call on the commit path. Blocks merge into higher tiers as material runs out, and decompress or search_context reads any block back at its original length. The background observer spends model tokens: it uses the session's provider and model unless you pick another, with up to 10 distillations at once by default.

Manage DeepSeek Harness agent presets (personas) from a Settings page: create one on top of a shipped base preset, edit a persona in place, lay it out in a rendered preview, and export or import it as JSON or Markdown. Edits to a hand-written profile patch are applied as text, so comments, quoting and loader expressions survive. Desktop app is the tested environment; creating from a base preset needs the yaml package that ships with DSH.

Per-workspace model packs: bind a named provider, model and reasoning effort plus the four subagent tier routes to a directory. Applies to new sessions in that directory; a session that has not run a turn yet also follows a pack switch, while sessions that already ran a turn are never changed.

Registers a battle-tested development delivery rulebook as a skill family - one core skill (clarification, delivery checklist, navigation) plus nine focused skills (layered architecture, evidence-driven debugging, minimal changes, impact-graded verification, migration feature parity, async UI feedback, sizing rules, CLI safety and cleanup, project notes). Only the skills matching the current task are loaded, trivial tasks (small reversible single-file edits) are exempted outright, so each session pays tokens just for the rules it actually uses.

Provider limits in the DSH sidebar footer: the remaining OpenRouter credit balance and OpenCode Go rolling 5-hour, weekly and monthly quota windows, shown as bars or percentages with a Remaining/Used choice, plus a Creator-mode request when the active provider is not supported.

A TRAE-宝-style personal assistant plugin for DeepSeek Harness: cross-session long-term memory, auto-extracted candidate memories, first-time onboarding, a dedicated assistant agent preset, scheduled daily digest, desktop notifications, and a live overview panel.

Windows-only stealth lyrics readout: reads the desktop Netease Cloud Music client through Windows SMTC (song, artist, play state), fetches its LRC and translation from the public Netease API, and renders the currently sung line directly beside the DSH activity line ("深度求索中，用时 X s") with font, size and color cloned from that line via computed style, so it reads as one longer sentence. The stopwatch-based progress freezes on pause and resets on track change; lyrics hide entirely when no activity line exists; an anchorText config retargets non-Chinese hosts; 1.0.5+ fails over between two Netease search endpoints when rate-limited. Needs the Netease "SMTC" toggle enabled.

Reasoning-effort slider that replaces the composer model seat: one rail over the levels the current model declares (off/low/high/max on DeepSeek), with a 241-frame portrait that evolves frame by frame as you drag and a per-model choice that survives restarts. Uninstall restores the shipped picker.

Adds a Print action next to Copy that opens a preview and prints the assistant message.