Plugins

Site bookmark manager for DSH: a folder tree, multiple links per site, tags and filtering, with per-folder views.

Shared runtime for the Personal Workbench: local SQLite, domain registration, agent tool, HTTP/SSE and opt-in cloud sync for the Tasks, Tags, Sites and Storage plugins.

Derives the index from your Markdown memory files on every turn instead of keeping a hand-written index table, so renames, splits and deletions never leave a stale entry. Retrieval ranks by library tags, section tags, title tokens and body text, then injects the matching line ranges rather than whole files. Writing is model-proposed and user-confirmed through a sticky note; entries are appended after the target section's format is detected. Ships a structure panel showing each library and section with line ranges and token estimates, a starter template for an empty memory directory, and optional appearance controls.

Local jot-down notes for DSH: one Markdown file per note, stored under ~/dsh随手笔记 by default (portable home path, no drive-letter assumption). Redesigned card UI, trash bin (deletions restorable for 30 days), custom tag management with stable per-tag colors (create tags ahead of use, delete one everywhere), compact/comfortable density, adjustable font size and line width, flat quick-scan list, full-text search, Markdown preview, autosave, batch deletion, zero-dependency Office text extraction, and AI write-back from conversations. Heavily reworked from dsh-knowledge-lite; pick the old library via the in-app root picker to migrate.

Bookmark a whole assistant reply with the star on its action row, or bookmark any selected sentence with the icon beside the selection; pins are saved as Markdown to a local folder or an Obsidian vault, and the existing right sidebar lists, tags, searches, exports and jumps back to them.

Long-term memory for dsh with defense-in-depth hardening. Two tiers (project `.dsh-memory/` and user `$DSH_HOME/memory/`), each a plain hand-editable JSON file with bounded entries and tags. Writes are atomic (unique temp file + rename, previous revision kept in a bounded history/) and serialized across processes by a per-tier lock with stale reclaim; a disk-revision check refuses lost updates instead of silently overwriting another writer. Memory text is wrapped in explicit untrusted-data boundaries (forged wrappers stripped on write) to blunt prompt injection from remembered content, load-time validation rejects out-of-bounds stores, and symlink escapes out of the tier directory are refused with MEMORY_STORE_ESCAPE. No content-heuristic filtering and no confirmation step, so headless auto-consolidation stays intact. Install: dsh plugin --profile web add github:RocoFish/deepseek-memory.

Material library for dsh: collect a folder, file or note from a conversation, keep it in real folders with tags, browse it Finder-style, reopen the source chat's output in one click. Ships a materials_archive tool plus auto-filing rules: file a task's deliverable once, and every later scheduled run files the new documents by itself.

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.

Market-gap radar for DSH plugin developers. Splits the catalog into categories and topics, tags every topic empty, thin, contested or crowded, and ranks the emptiest directions first, so you know what to build before writing it; it also browses the whole market with copy-ready install commands.

Reference search for DeepSeek Harness: refkit_search looks up image, video, audio and text references across 24 open sources and tags each result with its normalized license and a use verdict for the intended use; refkit_rights re-checks one license against another use. Source keys and search tuning live on a settings page. 11 of the 24 sources need API keys; the use verdict is returned only when an intended use is passed. Runs on DSH ^0.1.7-rc.2 || ^0.2.1-rc.2 (tested on 0.1.7-rc.2 and 0.2.1-rc.2).

Local knowledge base for DSH: one Markdown file per note, with tags, hierarchical folders, full-text search, attachment opening via the system default app, zero-dependency Office text extraction, and citation into the conversation. Built for host DSH 0.2.0-rc.2; on Windows the default root is E:/dsh-kb.

Skill browser and switchboard for the DSH web GUI: a sidebar entry lists skills from configurable source directories as cards, clicking a card opens the full SKILL.md, and the per-card switch writes disable-model-invocation and registers or removes that skill in the agent skill catalog, so the model can load exactly the skills that are switched on. It also recognises WorkBuddy expert-team folders and shows avatar, tags, quick prompts and members, registering its lead prompt as a skill when switched on. Source directories are editable in the panel.

Coding retrospect distilled memory for DSH: distills rejected solutions, pitfalls and constraints from each turn via a local LLM, tags them into MemOS (type:rejected_solution), and boosts weighted recall on coding tasks via agent/pre-step.

Checks git commit messages against the 7DGroup convention (Chinese type tags, length and punctuation rules) before committing, as a client-side guard before GitLab pre-receive hooks.

Manage and organize your agent skills by category, tags and collections in DeepSeek Harness, with a settings panel and a slash-command folder picker.

Attach tags to sessions and search sessions by tag in the Web settings panel.

Colored tags for sessions, shown on the sidebar row and in the conversation header, with preset colors, pin-to-sidebar, hover-remove and drag-to-reorder; tags persist in a Host-side storage domain and survive restarts.

Notes, to-dos and dated events kept as one JSON record per item in the local DSH profile and shared across conversations. Three tools list, create, update, complete and delete them, including adding and removing tags through conversation. A sidebar page groups events by day, remembers agenda navigation within the same Shanghai day, folds earlier unfinished events and lets you show completed events. An unfinished item with a due time raises an in-app reminder — including one missed while the app was not running. An optional WeChat push goes only to a target you have already saved, over the DSH-IM channel, and a successful send is recorded as the channel accepting it.

Registers a skill that turns a Bilibili video link into a structured Obsidian note. It fetches subtitles only with yt-dlp and never downloads the video, merges multi-part anthologies onto one continuous timeline, and writes front matter, timestamped jump-link bullets, a summary and tags into the vault.

End-of-turn voice summary: the model calls voice_say once before finishing a reply so the conclusion is spoken through Breeze TTS, with Chinese bracket audio tags (laughter, sighs, pauses) and optional performance instructions. Audio is synthesized host-side and played from browser memory without ever touching disk; the character limit is a writing target for the model, not a cut-off, so long lines still play in full. The API key stays in the credential store, and an optional daily Breeze check-in is included. Needs a Breeze API key and sends the spoken text to the Breeze service; its HTTP routes (including the settings route that stores the key) have no DSH access check.

Tool-surface governance for DeepSeek Harness: six scope-separated deny lists cut each agent's tool visibility through the host tools.restrict() API, unseen tool names wait in a pending gate, and dispatch_composed assembles a minimal surface from capability tags — a tasks array runs subagents in parallel, and a continued subagent keeps its original surface. Every decision goes to an append-only ledger that rotates into dated archives without ever losing history.

Drops text blocks that carry nothing but leaked DSML tool-call tags, then rewrites the resulting empty turn into a retryable error so the host's built-in retry re-issues the request; prose that merely contains tags passes through untouched.

Session archive with tags, notes and per-session metrics, plus a Plan-Do-Check-Act loop over the archived sessions that produces prompt and AGENTS.md rules and re-measures them on later sessions. Runs locally without calling a model.

Skill card gallery for DSH: every installed skill becomes a game-style card you can browse in four carousel modes, filter by category, source or tag, and edit — rarity, stars, five stats, tags and card art — while the card data lives in a side-car file, so SKILL.md is left untouched. Also creates and deletes skill folders.