Plugins

Technical-preview local graph memory for DSH and Codex: recalls owner-global and current-project evidence before each turn, writes USER and ASSISTANT records separately, and preserves project, session, actor, and source provenance for a visual atlas and cited knowledge pages.

Deep-collaboration writing agent for DSH: per-user writing-style learning from edit pairs, clarify→outline→section-writing→red-team→reader-gallery flow, project-to-report synthesis (lab report / product intro / survey / README), 41 MCP tools, and web selection-improve UI with in-app file preview.

Credential provider for DSH with password-protected AES-256-GCM storage, Argon2id key derivation (legacy scrypt v2 auto-upgrade), SHA3-256 integrity checks, and temporary runtime decryption.

Load Claude Code assets into DSH at runtime — skills, slash commands, rules, permission rules (deny/ask/allow), subagents and hooks from .claude/ plus Claude Code plugins (plugin.json / marketplace / skills / agents / MCP), via a memory-IR parse layer that keeps .claude as the source of truth with no writes back. Six installable packages: loader, skills, permissions, agents, hooks, mcp.

Chinese localization for DeepSeek Harness: translates all system prompts, tool descriptions, and runtime contexts into Chinese at runtime via the system-prompt/assemble waterfall (1788 entries, zero source modification, uninstall restores English).

Technical-preview local graph memory for DSH and Codex: recalls owner-global and current-project evidence before each turn, writes USER and ASSISTANT records separately, and preserves project, session, actor, and source provenance for a visual atlas and cited knowledge pages.

Neumorphic DSH Web theme with light and dark palettes, ambient lighting, material shadows, grain texture, frosted-glass surfaces, and micro-interactions.

Draggable 3D liquid-glass floating card for the DSH web GUI: DeepSeek API balance, cumulative spend/token stats (today/yesterday/7d/30d/all), adjustable glass/3D/color controls, and one-click top-up.

Build PowerPoint decks from a brief: plan and storyboard gates, per-page authoring through pptwise and ppt-master, a design-profile pass, page-image snapshots with a render audit, and a tool that hands those pages to an image-capable model for review.

Website cloning methodology skill: a real-source-first decision tree covering static sites, React/Vue/Next content sites and WebGL/Canvas-heavy fronts, mandatory verification of executable code quoted from second-hand AI analyses, and 13 zero-dependency Node scripts for CDP browser recon, asset/font harvesting, route crawling, sourcemap hunting and pixel-diff auditing.

Adds a subagent_codex tool that delegates tasks to the local OpenAI Codex CLI, running the work in an isolated external process.

DSH Desktop enhancement plugin: it puts verifiable facts on the table. Constraints are injected only when relevant, so idle chat costs no extra tokens; injection is layered (session-stable text in the system prompt, volatile content as a tail snapshot) and the prompt cache stays valid. Discipline: each turn is classified (question / research / discussion / diagnosis / execution) with boundaries enforced - no file edits on question turns, no unsolicited fixes on diagnosis turns; citing code lines not opened this session, negative claims about unseen symbols, requirements that do not match the deliverable, and dressing up “I did not check” as “waiting for your decision” are all called out; replies are written for humans - a code, field name or id must be explained in plain words the first time it appears. Memory and knowledge: 75%/90% context warnings plus an exported session recap (say "continue" in a new session); project knowledge accumulated across sessions by working directory, auto-captured and deletable by id. Inspectable artifacts: task contract / change record (auto-logged, unverified items flagged at delivery) / hypothesis record (conclusions must pass an evidence gate) / design decisions. Dashboard and personas: routing, trigger hits and block assembly are written to lume-metrics.jsonl and queryable via lume_metrics; personas are distilled from chat logs, novels, scripts or character sheets, with long-term and temporary memory, a memory star map and persona card export/import.

DeepSeek Harness plugin that aggressively saves tokens without affecting task quality: lossless head/tail truncation of tool results (default), grep-result folding, pressure-triggered dsh-compaction-basic delegation, and an LLM-summary CompactionEngine subclass for the aggressive mode. Saves 97.6% of input tokens on an 8-scenario synthetic suite, with all reductions routed through dsh event flow so model-visible = logged stays intact.

Prompt enhancement for the composer: an "✨ Enhance" button rewrites your draft prompt into a clearer, more structured version with first-principles thinking, grounded in session context, memory and recent conversation; asks clarifying questions in a card when key info is missing, then previews the result before applying it to the input.

Detects which conversation model an account is being served: launches concurrent throwaway probe conversations, scores each streaming chain-of-thought by how its paragraphs open, and cancels probes that read as the older model within seconds. Includes an offline self-check that replays 13 hand-labelled chains-of-thought through the same classifier.

Vault button beside the DSH composer: save keys, servers, phone numbers, addresses and URLs, auto-collected from conversations, password-protected with WebDAV backup; click to insert, Ctrl+click to send.

Web Audio lifecycle cues for DeepSeek Harness: ten configurable events, three low-noise defaults, per-event custom sounds, and no shipped audio assets.

LAN / remote-access control, frpc NAT tunneling, a panel password gate, and mobile UI adaptation for the DSH Web GUI.

12 tested engineering skills (git commit, code review, test-first, plugin dev, debugging, changelog, PR review, task breakdown, and more): every SKILL.md passes a format validator aligned with the official skill rules plus an isolated load smoke test.

A lightweight, general-purpose attachment Dock for DSH: paste, drop, or choose images, PDFs, documents, archives, code, and other files, review and manage them before sending, then pass them to the Agent as local paths. It gives non-image files a consistent attachment entry point.

Network proxy settings page for DeepSeek Harness: set a global HTTP(S) proxy with bypass hosts, or give each LLM provider its own proxy or a direct connection; changes apply on save without a restart.

A Witch on the Holy Night (Mahoutsukai no Yoru) visual-novel theme for the DeepSeek Harness Web UI, with original-style menus, three characters with personas, day and night playlists, and an ending sequence. The repo ships no assets of its own; the 380 files / 263 MiB it drives (136 sprites, 208 UI parts, 13 backgrounds, 23 audio) must be recovered from your own copy of the original game, and the screen stays largely empty until then. The 23 audio files are a header strip away from working OGG; the 357 images use proprietary .mzp/.cbg formats, so the per-file mapping guide offers two routes and one of them expects you to supply your own decoder.

TypeSafe Jev as a first-class Cordis service and three model-visible tools for DeepSeek Harness — offline by default, egress disclosed, never default-on.

Startup splash for DeepSeek Harness: an animated avatar with aurora, stardust, pointer parallax and a module-assembly entrance, plus the same image as the main-UI wallpaper; both images can be swapped from a settings page that previews the animation live.