Plugins

DSH Web plugin for managing sessions: archive/unarchive, guaranteed physical deletion (transcript files + ~/storages/*.json sweep), drag-and-drop workspace reassignment, and branch-based message editing (edit/retry/reroll with reversible version timelines

Experience validation layer for DSH: zero-token collection of tool-failure/retry/search signals, AI refinement into verified skill books (three-layer verification), looked up at task start. Ships 10 trial skills and benchmark data (complex-task success 100% vs 60% bare).

Transparent multimodal routing for text-only models: every image in every model call is fully transcribed (verbatim OCR, data, uncertainty zones, injection-hardened) by your own multimodal understander, with focused re-look via vision_relook and automatic retry on image-related failures. No bundled endpoints, no borrowed logins.

Auto-retry failed API requests to keep DSH conversations going. Supports configurable status codes and error codes. 自动重试失败的 API 请求，让 DSH 对话不中断。支持自定义状态码和错误码。

Mojibake guard bundle for DSH: feature-based garbled-text detection before writes with review-then-release (an identical retry passes), plus pre-execution pwsh encoding audit (ANSI/GBK/UTF-16/Default), Windows and Chinese first.

Keeps a turn from closing on an unfinished reply: a stream guard turns silent truncation into an in-step retry, and a completion protocol asks the model to declare how each turn ended.

Read-only /btw side questions for DeepSeek Harness: forks the current session, injects the running turn's task, recent tool calls and partial answer so a mid-turn question has context, denies every tool call at execution time, and keeps a per-session topic history with follow-up, stop, retry and delete.

Sends desktop notifications when a DSH turn finishes or fails, and sends 继续 automatically after the model-retry budget is exhausted.

One-step local model onboarding plus a light harness for DSH. On load it probes loopback and private-network inference endpoints (llama.cpp, Ollama, vLLM, LM Studio), reads each model's context window and output budget from the backend, and registers the provider through settings, so a local model becomes selectable without hand-writing config. It then trims only that provider: masks the tool list through tools.restrict, drops named system-prompt sections and memory contexts, swaps the injected instruction block for a short one while preserving its first 80 and last 96 characters, and truncates tool-schema descriptions without touching structure or required fields. Cloud providers are returned as the same assembly object, untouched. The injected block is re-read from disk on every assembly, so editing the file takes effect on the next request with no reload. The panel in the plugin detail page is a live operations view: it shows whether the takeover is active and by which hook, how many characters the last assembly was trimmed from and to, and which providers are online (probed concurrently, with a real one-shot completion reporting latency and reply text per provider). If a local backend is not running, the panel offers a one-click launch: it probes the port first (never starting a duplicate), then starts the matching server (ollama serve, or a custom command you supply) detached and polls until the endpoint answers, clamping the wait to 3-600 seconds, reaping the child it just spawned if it never answers, and naming the paths it searched when it cannot find the binary. A stop button kills backends this plugin started (SIGTERM, then SIGKILL) and refuses to touch a server somebody else launched. When a model has just been downloaded and no backend exists yet, the panel can also scan the machine for .gguf files with hard caps (300 models, 20000 directory entries, Downloads only one level deep), start llama-server on the chosen one (skipping the .mmproj. vision projection and reusing a server that already answers), register the provider under the model id the backend reports, and run one real completion to prove it works. Before spawning anything it does a fit check that counts only the part of the file the GPU cannot hold: hostBytes = fileSize - VRAM, compared against max(total RAM - 2GB, free RAM). A file far larger than RAM is refused outright; a heavy host-side remainder is refused unless you confirm; the same remainder under mmap is only a warning, because the weights land in reclaimable page cache. A load-mode selector (auto / mmap / none / mlock / mmap+mlock) maps to llama.cpp --load-mode, and the default is whatever the engine picks. From 0.4.2 everything above is derived from the platform instead of assumed: home and data directories come from os.homedir() and the OS conventions (LOCALAPPDATA on Windows, ~/Library/Application Support on macOS, XDG_DATA_HOME elsewhere); backend binaries are looked up in the official install locations of each OS and can be pinned with OLLAMA_BIN, LLAMA_SERVER_BIN, VLLM_BIN or LMSTUDIO_BIN; model discovery walks DSH_LOCAL_MODELS_DIR, ~/models, ~/llama.cpp/models, ~/.cache/llama.cpp and the LM Studio data directory, with the WorkBuddy layout demoted to one candidate among them. VRAM is asked from nvidia-smi, then rocm-smi, then sysctl hw.memsize on Apple Silicon (flagged unified, so the same bytes are not charged twice), and any flip to a unified-memory budget is handled explicitly. Before building a command line the plugin runs llama-server --help once and keeps only the flags that build actually has, so older engines get --n-gpu-layers 99 and no KV quantization while newer ones get auto plus q8_0; anything it cannot probe is passed through rather than dropped. LM Studio is started with lms server start --port N rather than being lumped in with the engines that need a model argument, and vLLM gets its own --model example instead of llama.cpp syntax. Windows, macOS and Linux are all exercised in CI, including a windows-latest and a macos-latest job, because a cross-platform claim that has only ever run on one OS is not a claim. Two of those suites exist specifically to catch this machine's blind spots: one replays the whole suite with process.platform rewritten to each OS, and one deletes every advantage this development box has — no nvidia-smi, no rocm-smi, no sysctl, an empty PATH, a home directory with no models — and requires graceful degradation rather than a crash. That last one caught a real class of bug: several assertions hardcoded 'lms.exe', which cannot pass where the executable has no suffix. Measured on the reference machine: a 16.9GB 27.2B MoE with 14.5GB of VRAM holds a 285-691MB working set and sustains 76 tok/s after a 1000-2000 token warm-up, so file size alone is not a fit signal. From 0.4.3 the scanned models are shown as a table rather than a dropdown (one row per gguf, with its size and any fit warning), so you can see what is on disk before choosing. From 0.4.4 the delete button on the provider table removes the provider entry, not the weights file on disk. That distinction matters: what people actually want to clear out is the dead entry - one pointing at a port that never answered, one registered twice for the same port, one whose model file was deleted long ago - not a 15.8GB gguf. Deleting a config is undone by importing again; deleting a gguf is not. The removal goes through settings.mutate with an unset op on \['providers', id], because update merges layers: writing null there leaves a null provider key that the loader reads as a broken config rather than an absent entry, and only mutate/unset is implemented as Reflect.deleteProperty. The result is read back afterwards, because unset on a key that comes from a lower layer (the profile cordis.patch.yml or a home patch) does not delete it - it restores the inherited value, so the call succeeds while the entry stays. When the entry is still there the panel says so and points at the file to edit instead of reporting success, and a DSH whose settings service predates mutate() is told plainly rather than silently ignored. Deleting takes two clicks and clearing the confirmation is as easy as clicking another row; on success the online and baseURL caches are cleared and the tool surface is recomputed, since the provider list changed. The gguf itself is never touched by this plugin - the scanned-model table is display and selection only.VRAM reporting is driver-dependent. From 0.4.5 model discovery also reads Ollama's own storage: its blobs are named sha256-… with no extension, so extension matching could never list them. Weights are now accepted by extension OR by GGUF magic header (only dot-free files of at least 8MB, capped at 600 header reads per scan; obviously-not-weights extensions like .json/.md/.safetensors/.part are never opened). Ollama models are resolved from their manifests: the model layer is picked by mediaType application/vnd.ollama.image.model (never adapter/mmproj layers), the human-readable name is restored from the manifest path (…/qwen3/8b becomes qwen3:8b, …/latest stays bare), tags sharing one blob merge into a single row, and the metadata kv supplies name, quantization and context length. Adopting an Ollama model does not start a duplicate llama-server over the same weights: it confirms ollama serve answers on 11434 (launching it if not) and registers the provider under the tag the endpoint itself reports — the blob's sha256 name would never match anything, since 11434 lists qwen3:14b, not sha256-a8cc…. Two real-machine-only traps fixed on the way: existsSync('<file>/') returns true on Windows, so manifest-vs-directory branching now goes by parseable content instead of trailing-slash semantics; and the quantization label table had been copied from a stale ggml enum generation, labelling ggml_type 15 as Q5_K_S while ollama show reports Q4_K_M. Same-blob dedup is by (dev,ino), keeping whichever entry has a name. Verified on the reference machine: the scan goes from 1 row to 4 — qwen3:14b (8.64G) and qwen3:8b (4.87G) now appear with Q4_K_M badges — in 4ms; npm run verify exits 0 (launcher 57, portable 35, panel-host green including the linux/darwin simulated-platform runs); the new panel-host suite builds a real miniature Ollama store in tmpdir and points OLLAMA_MODELS at it, so results no longer depend on whether the test machine happens to have Ollama installed. From 0.4.6 the panel shows each model its real name instead of its file name, because some download tooling stores every weight under one generic file name, leaving the real name only inside the GGUF header. The scan now reads each file GGUF KV section - not a fixed-size head buffer, because tokenizer.ggml.tokens (hundreds of thousands of strings on a large MoE model) and the chat template push general.file_type past any window; and GGUF string arrays are interleaved (u64 length plus payload per element), so treating them as a length table walks the cursor off the rails and produces astronomically large reads. Reading and skipping are accounted separately: key names and types are always read, string values over 256KB and string arrays are only walked past without copying, non-string arrays over 64 elements are skipped outright, and a cursor past the file size stops the read, because an out-of-range position means the cursor is wrong rather than that a limit should be raised. The displayed name goes Ollama tag, then general.basename plus size_label, then general.name, then the file name, with the full string kept for the hover text since it usually carries internal build numbers; mmproj files are now identified by general.type as well as by name. A short read (truncated file) no longer feeds half a value into the parser. On the reference machine, reading one file header parses 44 keys in about 16ms, and a whole scan takes about 31ms. From 0.4.7 the smoke test no longer fails the first time and passes the second. The old code gave both request attempts one shared AbortController and one shared 30-second timer, so the total budget covered both rounds: a cold load of a large model burned it in round one, the abort left the signal permanently aborted, and the second round (the Ollama-style reasoning_effort: none retry) could not even be sent - which is exactly why the first click returned This operation was aborted and the second succeeded. Now each attempt gets its own controller and timer, a timeout retries once automatically (retryOnTimeout, on by default) because a first-request timeout almost always means the weights are still being read from disk, and non-timeout errors (refused connection, HTTP error) are not retried at all. Timeouts report themselves: the message says the wait exceeded smokeTimeout, whether a retry already happened, and points at the new smokeTimeout setting (seconds, default 60, clamped to 5-600) - port responding is not the same as tokens coming out, so it is a separate knob from launchTimeout. Unrelated: an example naming one specific model was dropped from the panel copy, since not every machine has that model downloaded, and the explanation now describes the mechanism instead.

Prepares files before read, read_image and the vision tools open them: downscales images over 2048 px or 8 MB (one retry at 1280 px within the same turn), converts BMP/TIFF to PNG, relabels mislabeled uploads by sniffing their bytes, and turns docx/xlsx/pptx, text-layer PDF, csv/tsv and html into Markdown with no external programs.

溢出兜底与续命：上下文超限、内置压缩放弃后自动再压一次并重试；仍失败时把原会话整理成交接包，在新会话接着聊。不换引擎、不加界面；续命默认显式触发，原会话只读。 · Overflow fallback and session revival for DeepSeek Harness: one last compaction retry, then continue the dead session in a fresh one with a handoff. No engine swap, no UI, o

In-process crash guard: takes over the fatal-error handlers that would exit DSH, names the responsible plugin from the stack, and disables only that row in the home patch layer, leaving the profile and the provider config inside it untouched. The crash page offers retry, keep-disabled or uninstall, and opens a temporary safe-mode session when no culprit can be named (CLI and web launches only).

Rejects foreground waiting in DSH agents: waiting commands (sleep, wait, tail -f, watch, ping without -c, infinite loops), blocking wait tools (wait_agent, job_output with wait: true), progress polling (the same check command repeated within a window) and commands that need human input/confirmation (sudo without -n, ssh without BatchMode, login flows, interactive editors, docker -it) are refused with an actionable retry to run in the background or end the turn. Configurable budget and polling window, message language follows the system (zh/en), zero runtime dependencies.

Retry failed model requests automatically: intercepts every request error and replays the same step after a 2s / 5s / 10s / 20s backoff ladder, so gateway blips and rate limits no longer end the turn.

GLM Coding Plan usage quota ring in the composer dock - 5-hour and weekly credit windows with level, auto-refresh and one-click retry. Requires a GLM Coding Plan subscription and key, plus your BigModel organization and project IDs (the team quota API returns empty without them).

Adds a native Restart DeepSeek Harness item to the Windows system tray menu and the account menu of the DSH desktop app: the plugin prepares a patched resources/app.asar at startup, a detached helper backs up the original and swaps it in after a normal exit, the patch is re-applied after each desktop update, and either item relaunches the app gracefully on the next launch. Windows desktop only; other hosts report 409 not_desktop.

Button beside the composer that rewrites the current draft into a more executable prompt in place. Ships four built-in presets spanning light polish to aggressive expansion, accepts user-defined presets, and switches preset, prompt model and reasoning effort directly from the button without leaving the composer. The button appears only when the draft has content, shows a running timer while enhancing, and surfaces failures with a retry.

See your real ChatGPT / Codex subscription quota inside DSH. Reads /wham/usage: 5-hour and weekly windows, reset countdown, extra-usage balance, additional limit groups and the code-review limit; lists and redeems server-issued reset credits with a stable, retry-safe idempotency key; switches between multiple accounts; reads /codex/models to show the account's real model catalog, showing empty rather than a built-in list when unavailable. One-click sign-in from Settings (browser authorization or device code, with a paste-the-callback fallback); credentials stay in DSH's credential service. Model calls reuse DSH's built-in llm-pi-ai Codex route instead of claiming a provider route — it degrades to a pure quota console when another plugin owns it. A remaining-quota chip sits beside the composer.

Windows sandbox ACL-repair plugin for DeepSeek Harness (dsh) — fix the grantWrite failure (SetNamedSecurityInfoW Win32 5) and retry the command invisibly

Forward WeChat chat records into DSH sessions: a Windows share target captures the merged-forward ZIP, an in-GUI overlay picks the destination workspace and session, and the ZIP arrives as a file-attachment user message the agent unpacks and reads. Media entry extensions are normalized by magic bytes (WeChat ships PNG/HEIC bytes named .jpg), and failed deliveries preserve the batch for retry. Requires Windows 10/11 with Developer Mode on and the repo installer to register the share target.

Union IM bridge for DSH — an independent implementation uniting the capability sets of community peer plugins (xmanrui/dsh-im, MichengAI/dsh-im-connect). Drive your agent session from 15 channel adapters across 12 platforms (including email; WeCom/DingTalk/QQ each contribute two: bot + self-built app or notify tier) — WeChat (official iLink QR bind), Feishu/Lark (long-connection + streaming cards; the voice tier requires system ffmpeg), Telegram, Discord, WeCom intelligent bot & self-built app (signed callback), DingTalk (stream + AI Card and custom-robot notify), QQ official bot (plus notify tier), Slack (Socket Mode), WhatsApp (Baileys QR pairing; the dependency ships as an optionalDependency of the plugin package), Matrix, iMessage (macOS only), Email (IMAP/SMTP thread routing). The cross-version narrow waist auto-adapts settings gen1/gen2/gen3 across hosts 0.1.2-rc.1→0.2.1-rc.2 (12-tag enum; rc.2 verified by a real desktop-profile install with double read-back) using capability probing only, never version branches. In chat: a 25-command panel, tool-approval and question loops answered in the originating chat (deny-by-default, initiator-only, 5-min expiry), image/file input gated on host model metadata, deliverable file return, a delivery log with bounded retry (uncertain is never re-sent), proactive delivery and fragment composition. It also splits the native sidebar into self-owned Tasks / Channels tabs (standing down entirely when another plugin already owns the slot), registers an `im-concise` agent preset plus reply sanitisation that keeps GenUI fences and raw HTML out of text-only chats, renders `svg` fences to PNG images, guards model choice with an allowlist, and mirrors each turn's final answer to notify channels via `/relay` (addressed by account ID or alias, never guessed on ambiguity). Secrets are one-way into the vault and never echoed back. npm: `dsh-im-union`.

Bridges DeepSeek Harness to a Hindsight memory service using the same conventions as omp: one shared bank, automatic project-tag stamping, project plus global recall, plus tools for mental models, duplicate detection and retrying failed background operations. Requires a running Hindsight server (default http://127.0.0.1:8888).

Observes consecutive identical tool failures and optionally pauses before the next model step, preserving original results and successful repeats, with an in-chat report in DSH Web.

Recovers broken DSH turns: widens the product's own retry budget and renders the wait as a live countdown inside the same step, keeps a truncated output alive inside the same turn, and can open an opt-in continuation round when a turn has already ended; while the composer is empty it hands the send button to that continuation.