Plugins

IDEA-style Git workbench for the Web UI: commit graph with lanes and ref labels, log filters, commit details, split and unified diffs with per-hunk staging, a fine-grained zero-context mode, local changes with amend and sign-off, rename-following file history, per-line annotate, stash, branch operations, compare, tags, an interactive rebase planner and conflict resolution. Records which model produced a commit by stamping an Assisted-by trailer from the live session, and renders any existing commit trailers as chips, so AI-authored history stays traceable.

Checks a site safety-qualification ledger for training hours, the four required record elements, and special-operation certificate validity against the dates the person worked.

Three-layer memory for DeepSeek Harness — a small always-on core block plus on-demand topic files (profile, preferences, decisions, lessons), with source-annotated writes and supersede-not-overwrite history, so preferences and project constraints stop drifting between sessions. Local-first: no vector DB, no server, no outbound calls.

Rates how hard the current message is and shows the cheapest model and reasoning effort that still fits, above the composer, before you send.

JetBrains Database-style data source tree and SQL/Redis consoles with 15 agent tools for DeepSeek Harness. SQL needs Java 8+ installed, Redis does not.

Brain-shaped relationship graph viewer for DeepSeek Harness with a Markdown knowledge-base pipeline that imports documents and builds the graph. The viewer ships demo data and needs no Python; rebuilding the graph from your own notes needs Node.js 20+ and Python 3.8+ and runs from the command line.

Multi-agent Chinese debate simulator with eight debaters, live two-column discussion and draft notes, judges and scoreboard, single-side clocks, per-round ZIP archives, a training side with a configurable end condition (win-rate threshold or fixed rounds), and a Word prep package.

Plain-language trust dashboard for the dsh trust family: aggregates the nine channels (injection interception, write denials, egress blocks, custody-chain verification, tool-surface and prompt drift, leftover scans, forgotten compaction constraints, profile composition) into one right-sidebar panel with a single status light and event cards that translate engine jargon into sentences a non-security user can act on; the read and egress engines run from the package's own npm dependencies so those channels work even without the sibling plugins, and absent plugins simply show as not installed.

Compaction integrity: directive lines (never/must/必须 style markers) extracted from the compacted range are verified against the summary — verbatim or by content-word coverage — and every constraint the session silently forgot is reported and written to a JSONL ledger; pinned constraints from config are checked on every compaction.

Training Lens opens a local TensorBoard scalar dashboard with selectable metrics, explicit improvement direction, optional EMA smoothing and Chinese/English UI; requires an existing Python environment with TensorBoard.

GPT-SoVITS voice output for DSH: a speak tool and /voice command, a read-aloud speaker button under each reply, and a settings tab to switch voices, tune synthesis, manage the local service and one-click install a voice pack. Windows only; needs a local GPT-SoVITS checkout with trained weights and its Python environment.

Structured image recognition: local sharp measurements plus a cloud vision model return description, OCR with coordinates, object locations and fine-grained attributes in one call; a fingerprint cache makes repeated images return in milliseconds without spending model quota.

Pins a Chinese (zh-CN) thinking and answering constraint into the system prompt, and registers a per-step runtime reminder so the agent does not drift back to English after tool output. Zero runtime dependencies, single file; falls back to a second registration path and only logs a warning when the host API changes.

Turns a Minecraft Java save into a Blender cover or animation: scans the .mca region files for flat open ground, exports the terrain to OBJ, and drives Blender headless with MCprep materials, the BPS character rig and three-point lighting. After each render it reviews the frame with OpenCV for facing, ground contact, exposure, lighting and composition, then re-renders with adjusted parameters when a check fails. Requires Blender 4.2 LTS or newer and Python 3.10+ with opencv-python and numpy; the first run assembles the Minecraft assets with the bundled setup command.

A philosopher's way of thinking made executable: dialectics and practice-testing to govern hallucination and long-session drift, so done is a verifiable fact rather than a claim.

Gowin (高云) 1.9 FPGA toolchain control for dsh: 15 tools covering environment and license probe, .gprj/.cst project and constraint auditing against a real GW1N board pinout, gw_sh batch synthesis and place-and-route, pnr resource and timing report parsing, programmer_cli JTAG scan and flash, arbitrary gw_sh Tcl, and the GAO on-line logic analyzer capture chain — plus a bundled gowin-fpga skill and three field-note references. Pure JS, zero runtime dependencies.

A floating panel that lights up which HowNet semantic primitives the current answer touches, as a brain-shaped node graph, EEG-style channel traces and a composition card.

Two games inside DeepSeek Harness where the model supplies intent and a local engine owns the rules: a twin-brain snake that periodically draws a strategy card, and Chinese chess where the model only comments on a move picked from locally generated legal candidates. Both keep running when the model is slow, offline or out of quota.

Installs a self-evolving development-constraints rule library, exposes it to the agent through a constraints tool and a once-per-session reminder, and audits AI token cost and context usage.

Zagent — unified three-role self-improvement loop with three operators. Data-RSI (daily cross-sectional cleaning: winsorize → cross-sectional standardization → median fill, plus experience extraction), Harness-RSI (evolves system prompt / memory / skills), and Model-RSI (evolves alpha training recipes via framework-native walk-forward with 2-day gap and multi-seed). Offline one-command run, GPU auto-detect. Ships one skill.

Colors DSH sidebar Session rows by their official status, in an Evangelion-flavored palette: a blue left rail for a row awaiting interaction, red for the session open in the main view, purple for finished-but-unviewed, and a static rainbow title for pinned rows. Every colour is mode-scoped — dark keeps the vivid stop, light takes a darker variant solved to clear 4.5:1 against the light wall — and the split reads DSH's own dark marker. Reads official Client services only: it owns no state, never polls, and ships no host-side half.

Governs the persistent information and project constraints an agent works with - project orientation and scope, information and document lifecycle, and persistent workspace mutation - through one additive system-prompt section plus deterministic runtime checks.

OpenCode Go quota in the sidebar foot under the workspace list: one cell for the most constrained window plus a hover popover with the rolling 5-hour, weekly and monthly percents, each with a usage bar, an elapsed-time marker, the remaining share and a reset countdown, refreshed every 60 seconds from OpenCode's official usage API using the key already in the DSH credential store.

Wallpaper-aware readability repair for the DSH Web GUI: it lifts the three low-contrast secondary text tokens with color-mix plus an adjustable text halo, adds content rules (bold, permanently underlined or recoloured links, tinted headings and body text) scoped to conversation Markdown only, offers adjustable rainbow-edge and gradient-fill heading effects, loads custom fonts from the plugin folder or the DSH home, and ships a draggable WYSIWYG preview with a WCAG contrast badge.