Plugins

Contract clause work: build the clause tree, map it against a 12-key playbook, register the playbook and export the unresolved report — 5 `legal.*` tools.

Multi-round deep research in the DSH sidebar: each round fans sub-questions out to parallel sub-agents, then continues from where the last round stopped and writes a cited report.

Whole-disk file search on Windows: ask for files or folders by name, path, extension, size or modified date and get matches from the local voidtools Everything index in milliseconds, with full paths, sizes and modified times. A second card reports whether Everything is reachable and how many items it indexes. Ships the official Everything SDK, so no es.exe or HTTP server is needed; Everything 1.4+ must be installed and running (the plugin can start it).

Read-only knowledge gateway over a local Markdown vault, in four tools: `wiki_recall` returns the matching section in full with page-level citations plus vault health checks and a page skeleton for writing a good answer back; `wiki_context` returns one evidence pack with per-page excerpts, a support verdict and an explicit list of gaps; `wiki_verify` re-checks a drafted answer against that pack and reports claims the evidence cannot back, including unsupported numbers; `wiki_observe` reports how often answers needed correcting. The settings page and sidebar panel share the same query path as the model, so a search you run by hand cannot disagree with what the model was given. The three knowledge-gateway tools (`wiki_context`, `wiki_verify`, `wiki_observe`) need a local Python 3 bridge — `bridge.python` plus `bridge.toolsRoot` containing `wiki_bridge.py`; without it they return `E_PYTHON_UNAVAILABLE`.

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. From 0.4.8 adopting a plain .gguf no longer registers an unrelated Ollama model. The port field used to follow the last click rather than the file: selecting an Ollama entry moved it to 11434, selecting a plain .gguf did not move it back, so a 11434 left over from an earlier click stayed. The server accepted it as given, probed the port, found Ollama answering and borrowed it (the skipped-launch log line about the port already answering is this path), and then the model id - derived from the file name - could not be matched against Ollama's /v1/models, where the fallback was to grab the first entry in the list. Embedding models tend to sort first, so the adopt registered a model the user never selected. Now the port follows the file: a plain .gguf landing on another engine's port (ollama 11434, vLLM 8000, LM Studio 1234) is moved to 8090 and the panel says so. Only ports belonging to a known engine are overridden - a hand-typed 9999 aimed at your own llama-server is left alone. The model-id fallback no longer guesses either: it takes the only entry when there is exactly one (a single-model backend may legitimately report a different id than the file name) and otherwise reports an error listing the ids it did see, rather than registering a wrong provider. From 0.4.9 the panel can no longer vanish. It used to be possible for the whole section to disappear from the plugin detail page while the host stayed up, the HTTP bridge kept answering, and no crash log appeared. The cause was a scope error: a helper was declared inside one component and referenced from another, which is valid JavaScript and passes a syntax check, so it only threw at render time - when the user pressed the scan button and React ran the state updater. The host wraps every slot entry in an error boundary whose fallback is an empty anchor, so the throw was contained silently: the panel was not uninstalled or crashed, it was replaced by an empty div. The helpers now live in the module scope both components share. The client half also gained its own test - a minimal React stand-in loads the real panel module, really calls the components, really runs the state updater at render time, and presses the scan button, so this class of error can no longer reach a release.

Adds a prompt-refine button to the DSH composer and, when auto mode is on, refines the prompt before the agent starts by rewriting the text with recent session context through a side-channel model call, asks on the native question card when the rewrite cannot be decided, reports each stage as a command card in the chat, and keeps the original text when analysis fails.

Detects Minecraft mod and plugin projects (Fabric, Forge, NeoForge, Architectury, Paper) and runs their Gradle client or server task from the session header, with launch logs in a sidebar tab. One-shot tasks such as data generation and GameTest report a pass or fail verdict when they finish.

Pipeline for scanned textbooks: OCR at roughly 5x the ONNXRuntime speed (requires Python 3.10+ with pymupdf, rapidocr, openvino and numpy), then a structure report and a per-chapter reading plan before you open the PDF.

Persistent cost ledger for the DeepSeek Harness: every priced model call lands in a monthly JSONL file; /ledger renders monthly reports, /ledger-export writes CSV, and the agent can query its own spend via ledger_query.

Conversation lock for the DeepSeek Harness sidebar: every conversation row gets a lock button, and a locked conversation blurs only the conversation area behind a password box while the sidebar stays usable. Re-locks on window blur, tab hide, page hide, or whenever the open conversation changes, and never pauses a running turn - the veil reports that background work continues. The password digest, lock flags, and preferences live in <DSH_HOME>/session-lock/state.json behind two same-origin routes guarded by a custom header, so clearing the page localStorage unlocks nothing; PBKDF2-SHA256 with a random salt, and a 30-second input lockout after five wrong attempts. Recovery is opt-in per machine: a Windows login password (enabled only after a real .NET PrincipalContext ValidateCredentials check against the local machine, so Microsoft accounts and passwordless machines keep it off) and a cooling period chosen while creating or changing the password (1/6/24/72/168 hours; expiry clears the password and releases every lock, and the correct password cancels the countdown). A settings section covers status, changing the password, lock new conversations, hiding locked titles, a three-position archive-restore policy (all / locked only / off, switching needs the password), lock everything, and unlock everything. Dialogs use the native dialog top layer, so a dialog opened from Settings is never covered by it. Requires DSH 0.2.0-rc.1 or newer; no build step.

Wire up Feishu/Lark from a four-step wizard that discovers your chats instead of asking for a 32-character id, creates the app by scan, and runs a nine-item live self-test that names the exact scope to enable when a check fails. It drives a companion MCP server of 25 tools, including ones that move local files into a chat and pull a message's attachments back to disk. The MCP server is a separate install: take dsh-feishu-mcp-<version>.mjs from the dsh-feishu Releases (or npx github:yinsang0910-star/dsh-feishu-mcp); until it is in place the self-test reports MCP as not mounted and no mcp__feishu__* tool exists.

Bundles a deep-read paper library and the paper-deepread-html skill: a sidebar panel to browse interactive reading reports, upload PDFs into a managed library, deep-read them with an agent (or through an optional site auto-read worker), and manage or rebuild the library; ships the paper_deepread, paper_deepread_uploads and paper_deepread_reports model tools. Deep-read/ingest and report builds require Python 3 (the bundled skill prepares a PyMuPDF venv on first use; index rebuild falls back to plain python3).

Adds a Continue button to the composer dock and resumes automatically when a turn is cut off by a network drop or an upstream output-limit cut. A resident link monitor asks every 10 seconds whether the network is back, because the link returning fires no event, and a turn that completes normally counts as proof the link works. The click goes through the host submission pipeline — beginSubmission plus session.prompt — so the message carries real user attribution instead of faking one, and a genuine refusal is reported rather than retried. On startup it also scans recent session logs for both fingerprints of a host that left mid-turn: an unclosed turn, and a turn closed as disposed, which is what a normal shutdown writes. A turn the user stopped by hand is never resumed.

Unofficial port of Ponytail 5.3.3 (upstream: DietrichGebert/ponytail, MIT; not affiliated) for DeepSeek Harness: minimal-code rules, session modes, native settings, a composer status band, and a reported-usage dashboard. It cannot be installed alongside another DSH Ponytail adapter — remove @mengyuly/dsh-ponytail first.

Annotate elements, blank positions or dragged regions in the DSH right-Sidebar Browser tab; each batch is saved as screenshots plus report files and attached to the next message as text. Does not replace the official Browser, Chat or Layout providers.

Static-context audit for a DSH home: a six-layer baseline (rules, memory, skills, tool surface, sessions, KB), broken-path checks for a markdown capability map, memory duplicate/stale/over-limit reports, SKILL.md description compression, and tool-surface plus cold-start static-context estimates recovered from session logs. Four read-only DSH tools; writes require --apply and are backed up first. Requires Python 3.10+ on the host (stdlib only; all four tools shell out to the bundled CLI).

Registers the workbuddy model provider by mounting dsh-llm-workbuddy and supervises the local wb2api gateway, then bundles two protocol bridges onto that same gateway: Responses for OpenAI Codex CLI, and Anthropic Messages for Claude Code and Claude Desktop, which see the gateway models under generated claude-wb-… ids. Skips work that is already running or already registered, and reports each part through a status tool. Requires a local wb2api gateway and a WorkBuddy account.

Qoder supplier plugin for dsh-router, covering both the CN and international regions. Uses a personal access token for authentication, signs requests to the Qoder cosy gateway and converts upstream SSE to OpenAI SSE; the model list reports the upstream display name alongside the key, while account selection and cooldown stay in dsh-router core.

Token spend panel for DeepSeek Harness: today and 7/30-day totals, natural month/quarter/year sums, cost split by cache-hit input, cache-miss input and output, per-model and per-session detail, daily and monthly budgets with a projected day-end spend, peak/off-peak pricing, CSV and JSON export, and a zero-dependency CLI report. Also prices GLM (Zhipu) and Kimi models shipped through third-party routes. Maintained fork of olimc2016/dsh-token-meter-panel (MIT, 0.6.7) with a rewritten panel; it uses its own row, route and slot ids, so it installs alongside the upstream entry rather than replacing it.

Route implementation work from the main model to executor models and collect their reports.

Auditable mathematical proving for DeepSeek Harness: a local stdio MCP server with 23 machine-checking tools, a bundled workflow skill, and the /prove and /audit-proof commands. proven, refuted and inconclusive are never conflated, and unproven steps are reported instead of hidden. Needs Python 3.10+ with pip install -r requirements.txt on first run; the 23 tools are unavailable until then.

One prompt row carrying two independently switchable halves. Identity: a fixed self-reference restated at the head, at the tail after every other section, and in the runtime snapshot, with an instruction to treat contradicting claims as stale cache. Discipline: ten rules, each paired with the condition a reviewer checks — external evidence before claiming done, change face equal to review face, correctness from named call points rather than timing, one source of truth, extension through the registered seam, runtime validation only at real boundaries, loud failure instead of silent skips, orthogonal outcomes reported separately, tests through the delivered entry point, and answer shape (conclusion first, length set by the change).

DeepSeek Harness shortcuts on your terms: a later assignment overrides an earlier one, displacements are reported, and even the composer's send/newline/complementary keys are rebindable.

Action cards: the agent proposes a concrete action (create a file, edit a value, run a command), you edit it and apply it in one click, and the Host executes it and reports back.