Plugins

Four filter chips on their own row at the foot of the sidebar, each carrying its match count beside the label: running sessions, sessions waiting for your approval, plan review or answer, sessions with something unread (a turn that finished unseen, or newer activity the panel was not opened on), and sessions updated since local midnight. A chip opens a popover listing its matches grouped by workspace, and a row opens the session. The row ends with a DeepSeek service status dot (green operational, yellow partial degradation or maintenance, red open incident, grey when the feed is unreachable) that subscribes to the status Atom feed once a minute with an ETag conditional request and shows the current finding, the refresh time and the three newest incidents on hover; clicking it opens status.deepseek.com. A separate panel adds an all-sessions chip, a more menu for the last day, three days, week or thirty days, and a manage mode that searches sessions by title or path, multi-selects them, lists the Markdown digests it wrote to the session-digests folder of the DSH data directory in a resizable, full-screenable right-hand column that previews them in place, and either moves the selected session directories to sessions/.trash or deletes them permanently after writing the digest and asking for a second, explicit confirmation. The session you have open cannot be selected. Every panel row also carries a star (shown on hover, filled once starred) and the panel adds a favourites tab that lists only starred sessions, so a long list can be triaged without adding another chip to the sidebar foot.

Drives a plugin page in the running DeepSeek Harness UI — opens the panel, reads real DOM and computed styles, clicks, types, waits for async effects, collects console errors, then returns to the original conversation.

Turn code flows into source-linked Mermaid diagrams and interactive CodeWalkthrough explainers.

Indexes each chat of a workspace into a small topic graph every 4 turns (one model call, verbatim source kept under .context-graph/), appends up to 5 matching topics from the other chats to each message, searching a named day such as "yesterday" first, and adds a recall tool that returns the original text by topic, chat, day or turn. On and shared by default; /context-graph off and unshare opt a chat out, and stats, build, map and find inspect it.

Mem0-backed long-term memory for DeepSeek Harness with managed Platform, self-hosted server and local OSS backends; visual settings, automatic recall (from the first turn) and capture, editable memory summary, token budgets and weekly memory hygiene that only deletes memories this plugin wrote. Captured text is redacted; the memory summary is a stable system-prompt section (plugin:dsh-mem0:profile) and recall a separate mem0:recall block appended last, so the cached prompt prefix stays stable; recall and capture have timeouts (recallTimeoutMs, addTimeoutMs). With dsh-strata installed, recall size follows its strataBudget allowance, and the optional captureStrataFolds setting (off by default) also saves its fold summaries as memories.

Turns a session into a folder in the left sidebar: fork it into an independent branch — full copy, a compressed digest of its memory, or only the opening request — and browse the branches as a tree.

Context handoff sentinel for DeepSeek Harness: three behavioral recall tests (constraints / changed files / next step) run after every compaction or on /guard check, cross-checked against the session log, returning a green/yellow/red verdict plus an optional handoff document.

Shows the DeepSeek API-key balance as one ambient stat in the Harness composer row, refreshed every 60 seconds. Click the balance to set a local low-balance threshold; the amount turns red below it. The API key stays in the host and is only sent to api.deepseek.com/user/balance.

Config-driven lifecycle hooks for DSH: block or observe tool calls, and react to completed turns, messages, and title changes with shell commands.

Plot decision gate for novel writing: seven beat_* tools register each node that changes the story into a decision point with 2-4 structured options; author picks are written to a book-level ledger (canon + obligations) that later chapters inherit, beat_gate returns pass:false until every decision is made, beat_compile turns the picks into a chapter contract, and beat_verify checks the prose for missing changes, forbidden states and overdue obligations. Off by default; /beat on or the composer button enables it.

Undermind literature search inside DeepSeek Harness. The plugin owns the OAuth authorization-code + PKCE login the remote MCP server requires, refreshes the rotating refresh token before expiry, re-initializes the MCP session when the server drops it, and registers every tool Undermind tools/list returns as a mcp__undermind__* tool, alongside a permanent undermind_auth tool for status, login, logout and retry. Needs an Undermind account and a client id - Undermind accepts a Client ID Metadata Document URL (CIMD) or a pre-registered client id, and does not support dynamic client registration.

Turn a script into a narrated explainer video: text-card slides by default (fast and exact for tutorials) or AI-generated footage per segment (needs AGNES_API_KEY); 14 online neural Chinese voices plus 3 offline Windows voices, with a one-command sample pack for picking a voice by ear. Requires ffmpeg, and the bundled video-remix skill copied to <workspace>/.dsh/skills/video-remix with npm install run there for the online voices.

Compiles LaTeX with the TeX Live installed on the machine: one `latex` tool typesets a formula into an SVG, PNG or PDF and a whole document into a PDF (Chinese text switches to xelatex with ctexart), writes the file into the workspace and returns a Markdown reference to it. A system-prompt section asks the model to write `$...$` and `$$...$$` instead of code fences, and when no TeX Live is found a banner above the composer offers a one-click BasicTeX or TinyTeX install.

Drives Siemens NX 2406 and 2506 from DeepSeek Harness, plus NX 12: an executor that judges a journal run from its log rather than its exit code (NX returns 0 for a journal that swallowed its exception, and 0xC0000409 after a clean one), 16 typed tools over a resident batch NX, and a probe that keeps the three releases apart. Windows only; NX 2406 is the reference release, 2506 expected. Requires a local Siemens NX install and a free NX licence seat; run `dsh-nx install` once, and the 16 MCP tools stay off until `NX_ENABLE_MCP=1` is set (the skill route works without 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`.

Unified long-term memory hub: captures turns into L0 evidence, distills in the background into L1 facts / L2 scenes / L3 profile (depth configurable), recalls via hybrid BM25+fuzzy search, ships 13 memory_* tools, traceable lineage back to source turns, optional Hindsight backend, and a built-in memory panel with three entry points.

Long-term memory for DeepSeek Harness: distils durable user preferences and project facts after each turn, injects a memory digest into later sessions, and keeps a daily diary of habits. Adds memory_search, memory_save and memory_forget tools.

Rewinds a finished turn in place — a button next to your own message withdraws that message and everything after it from the model context, puts its text back in the composer for editing, and optionally restores the files the turn changed through the DSH filesystem service.

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.

Turns the composer's model menu into a two-level list (provider first, then that provider's models), shows the current model together with its provider name, asks for a reasoning effort after a model is picked, and adds settings switches that hide whole providers or single models from the picker.

Deep-sea tidal-blue theme for the DeepSeek Harness Web client with a single accent color and matte (non-glass) surfaces in light and dark modes. Same-token UI additions: a single-line dock HUD reading real session projections (context usage vs. window, TPS, cache-hit rate, elapsed time), per-turn subagent delegation cards showing token usage and an optional cost estimate from OpenRouter public prices, sidebar pagination with pinned sections and type-to-filter, and model / reasoning-effort / permission controls under the composer. Each addition can be toggled in the plugin settings.

Cost pill at the right end of the composer stats row, showing the estimated cost of the current turn and the session total; clicking it opens a two-column panel with input, cache and output spend for the turn and the session, dialogue turns, today's and lifetime spend, and this session's cache-hit rate and cache saving. Rates follow the official DeepSeek price table including peak pricing (double on weekdays 09:00–12:00 and 14:00–18:00 Beijing time, statutory holidays excluded), each message is priced at the time it settled, and prices.json reloads without a restart.

Adds a chevron row under New Session that folds the sidebar panel list in place, returning the height to the workspace and session list. The hiding is driven by one attribute on the page root plus CSS: no host bundle is patched and no row is unregistered, and plugin entries injected outside the panel slot fold too because they are matched by their own data attributes. The state is remembered per browser and the control hides itself in the 56px rail.

A Web3D energy orb decoration for the DSH Web GUI: a particle core wrapped in three mechanical rings that shrinks and falls silent while the session is idle and pulses in a double-beat heartbeat while a turn runs. UnrealBloom glow, draggable with its corner, position and size remembered, and three.js served locally by the plugin instead of a CDN.