Plugins

Focus board for DeepSeek Harness agents: durable, model-maintained notes in the session workspace that pin the objective, constraints, and decisions across compaction and sessions, with automatic context injection, archive on clear, and an optional web panel.

Endurance-coaching skill bundle that defaults to Intervals.icu for training data (activities, intervals, streams, training load, HRV, resting heart rate, sleep in-bed hours, eFTP) and uses COROS only for training-plan read/write and for fields Intervals.icu lacks (sleep score, recovery score), with iGPSPORT as a fallback for ride details. It turns that data into readiness gating, VDOT pace and heart-rate zones, periodized weekly plans, session reviews, race-day pacing and fueling, and weather-adjusted workouts. The bundle registers the endurance-coach skill in the profile, ships deterministic helper scripts, reads an athlete profile created by its init script, and requires the agent to state when data is unavailable instead of inventing it. Data comes through separately installed MCP servers — the Intervals.icu MCP server for training data, the dsh-coros plugin for training-plan read/write and sleep/recovery scores, an iGPSPORT MCP server only as a ride-detail fallback — and its bundled scripts need Python 3.10+; with none of them it works only from numbers you type in.

Bridges the official COROS MCP server into DeepSeek Harness with a self-contained local stdio-to-HTTP bridge and on-device OAuth login, exposing 34 COROS tools on the current endpoint for sleep, HRV, resting heart rate, recovery, training load, activities and FIT files, plus workout and training-plan CRUD and scheduling.

Five-layer persona (L1–L5), event memory with FTS5 plus Dice retrieval, cognitive filter, skill cards with auto-frozen asset packs, millisecond reflex rules, three-tier safety degradation, a read-only brain star map, and an optional VPet-Simulator MOD bridge.

Nine single-ink print palettes — sage linen, kraft paper, lead type, cyanotype, risograph, CRT tube, phosphor screen, punch card, floppy disk. Each palette carries its own paper grain, type pairing, corner shape, section rule and a whale mark redrawn in its ink, plus a letterpress pass (paper tooth, sheet mottle, ink bite) that can be switched off. Every palette ships separate light and dark tokens.

JetBrains IDE bridge for DeepSeek Harness on Windows: installs as a DSH bundle, publishes the authenticated DSH URL, registers the open IDE project, relays the IDE MCP server, and serves a compact tool-window panel with streaming, approvals, Plan/Act, steer sending, session switching and archived-session filtering. The matching IDEA plugin is attached to the same release; your DSH agent can install it.

Autonomous product-polish loop skill: every round opens a fresh brainstorm from 3 forced perspectives, picks one small angle capped at ≤3 files and ≈100 lines, implements it, verifies with the repo's own tests, build or lint, and commits it as a single round commit; the round is logged to a per-project ledger, and ideas too big for one round go to an icebox. A red check ships nothing — the round commit is reverted immediately so the failed diff stays in history. Non-git targets are rejected before any work starts. Pure skill bundle: no MCP server, no runtime dependencies. Requires DSH 0.2.1-rc.2 or newer, below 0.3.0.

Zhinao (second brain) for DeepSeek Harness: Markdown-authoritative layered memory (L1 long-term model gated by user confirmation / L2 knowledge with mandatory sources / L3 recent status under daily quotas), a local recall ledger explaining every injection, lifecycle engine with weekly review and a Web panel. First-run wizard initializes a blank brain, adopts an existing Zhinao directory read-only, or fills in only missing skeleton files; existing files are never touched.

One click beside the composer turns a half-formed draft into a structured prompt the AI can act on: goal, constraints, output format and acceptance criteria filled in, placeholders only where a real decision is missing. Rewritten in place and never sent — undo with one click or Ctrl+Z. Four editable presets (general / code / writing / translation), optional conversation context to resolve references, and a model that either follows the session or is pinned to a chosen provider and reasoning effort. Requests survive session switches and typing while a rewrite is in flight; a result that no longer matches the draft is kept for retry instead of overwriting it. Pure JS, no build, no bundled dependencies.

Local LoRA and QLoRA fine-tuning in a panel: import JSONL, see a VRAM estimate before training starts, evaluate, and export an adapter. One-click Python venv setup, but you must supply your own Python 3.10+ and an NVIDIA GPU (6 GB or more). Verified end to end on Windows 11 with an RTX 3060 6 GB; the installer refuses non-Windows hosts, so Linux, macOS and other GPUs are unsupported for now.

Global cyber-terminal theme: a click-through full-screen Matrix code-rain overlay, phosphor-green UI styling, a sidebar one-click toggle, and adjustable speed, density, opacity and glow with three palettes.

ADHD-friendly focus soundscape docked under the DSH composer: five procedurally synthesised noise colours (white, pink, brown, rain, fan), a live-waveform orb, a 15-90 minute fade-out timer and a liquid-glass popover. No audio assets and no network requests.

内置 zk 的 DSH 知识库 / 日记插件：装好即用 —— 自动装 zk、建笔记本，给 agent 一组 zk_* 工具（搜索、读写、日记、双链），设置页里管路径、索引与自检。

Matrix-style binary (0/1) digital rain for the Harness interface: columns spawn only while a task is running, fall off screen when it stops, and leave a blank canvas; frosted-glass token layer with adjustable digit and panel opacity.

COROS Sport panel for DeepSeek Harness: sign in to COROS inside the panel, read the account, bound devices, recovery, fitness, recent workouts, daily health and the training schedule in one dashboard, and get a plan from the model with /coros-plan. The plan page writes a fixed four-week, 16-session plan to the account only after a two-step confirmation.

Lingshu startup injector: runs brain.py startup on every new agent session and injects persona / memory / todos automatically. Requires Python 3, and the Lingshu skill installed at ~/.dsh/skills/lingshu; silently does nothing if either is missing. Main repo also ships a bundled SKILL.md entry point.

A hard hat for AI coding: graded writes (L0 read-only / L1 workspace-only / L2 full) put privileged edits in a cage. Below the level it refuses to write and hands back a paste-ready block instead. Every path passes a raw-string traversal check plus realpath, so junction escapes are caught.

Distills daily chats into an updatable character card and a cross-session emotion module that trains the persona.

Three retrofuturist skins for the DSH Web GUI — a dusk train, neon rain and a sunset coast — with an in-page switcher and locally saved selection.

Private Windows DSH cold archives and guarded fresh-environment recovery. v0.7.0 adds precise browser-proof diagnostics and restrained, reduced-motion-aware UI feedback; strict dependency, transaction and verify/commit gates remain. Verified ZIP integrity does not certify private business recovery.

Sticker and emoji reactions for the DeepSeek Harness Web/Desktop chat: a composer sticker picker, agent stickers that render as images while the transcript keeps the pack tag, emoji reactions under the user bubble, and an emoji-rain effect.

Retro CRT phosphor theme: pure-black Matrix green, 40K amber-green or TFEL amber screens with bold glowing text, a code-rain boot animation and a BIOS-style control panel.

Ten full-canvas animated backgrounds for the DSH conversation pane — starfield, code rain, circuit, sword-fight, cat, girl, fighter jet, racer, snow and bubbles — one random theme per turn, with a text-column veil that keeps the chat readable.

Audit what context compaction actually lost. Reads the raw session event log and checks session-constraint survival (the standing instructions the user gave — do-not-X, must-Y, only-in-Z — and whether the summary still carries them; arXiv 2608.11242 measured ~17% survival across compactors), summary hallucination (paths, versions, ports and identifiers the summary invents that were never in the history it replaced), ghost references ("see above" pointing at removed text), token accounting (a summary nearly as large as what it claims to free), and unaccounted prunes. Four agent tools including a cross-session sweep that ranks conversations by how much compaction hurt them. Deterministic — no LLM, so a wrong answer is a rule bug with a name.