Plugins

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.

DSH plugin (fork of sj244/dsh-qq-bridge, MIT) that connects DSH sessions to QQ through OneBot/NapCat. Whitelist stays closed until filled in; @-mentions, nicknames and pokes always wake the model; other messages wake at a configurable probability. Cordis id remains qq-bridge. It keeps the upstream Cordis id (qq-bridge): uninstall sj244/dsh-qq-bridge before installing, the two cannot be installed together — and with an empty whitelist nothing wakes.

Adds per-model hide/show switches for custom (llm-pi-ai) providers to the DSH settings page. Hidden models disappear from the conversation model picker and are archived verbatim, so they can be restored at any time; disabling or uninstalling the plugin restores everything automatically. Built-in DeepSeek providers are not managed.

Kaipu Workstation — a cross-cloud, cross-platform console where agents collaborate on one workspace. Any agent runtime that speaks plain HTTP can take a seat, each with its own device credentials: project context is shared, files are kept per seat, and tasks rotate between seats with a trace of who did what. Workstation data lives on the server — this plugin is only the interface, so uninstalling loses no data and reinstalling restores it. The panel renders the persisted workstation state (no live stream yet).

Companion sidebar panel for the Kaipu site plugin: shows the built-in advisory team and seat status. Install together with kaipu-platform — seat data comes from the site plugin's gateway endpoints; a standalone install honestly reports "channel not ready". Removable: uninstall it and the site plugin stays fully usable.

MCP server console under Plugins: manage the official @deepseek-ai/dsh-mcp-client at runtime — add/edit/delete servers, enable/disable/reconnect, per-tool switches, live status and tool lists over SSE, mcpServers JSON import, and auto-migration of profile-YAML MCP entries; an uninstall leaves a native fallback block that keeps the same servers loadable.

Ponytail lazy-senior-dev ruleset for DeepSeek Harness: a 7-rung minimal-code ladder (YAGNI, reuse, stdlib, native, installed deps, one line, minimum), six skills (main rules, over-engineering review, repo audit, debt ledger, gain scoreboard, help), a /ponytail session command for lite/full/ultra/off levels, and optional always-on ruleset injection via config. Vendored verbatim from upstream ponytail (MIT), zero tools registered, clean uninstall.

Skills and MCP browser behind one floating ball with five tabs: Skills (native folder picker, two-level categories plus custom groups, tree or flat view, manual, keyword and AI grouping on metadata only, search, and per-card or whole-library enable/disable with timestamped backups), Call Ledger (counts, success rate, trend, ranking), MCP (live loader config, real Fiber states, per-server tools and call stats, enable, disable, reconnect, export to mcpServers JSON, plus scan-and-AI-fix for unconfigured servers), Market (MCP marketplace: the official MCP Registry plus 29 curated free servers, 19 flat categories with per-category counts, grid and list views with 3-6 cards per row, one-click install into cordis.patch.yml with automatic backup, one-click uninstall, every server shown with its own real icon, two-way instant sync with the MCP tab, and the last scroll position is recalled) and Quick Start. 130 skins sync across the three balls. Adapts DSH kernel 0.2.2-rc.1.

Loads Claude-style plugins into DeepSeek Harness (dsh): discovers skills, commands and optional SessionStart hooks from Claude plugin directories, with hot command reload and uninstall cleanup.

DSH 静态 bundle 插件：在设置页管理第三方插件——列表、启用/停用（持久化到 ~/.dsh/cordis.patch.yml）、卸载、检查更新、一键升级。A static-bundle DSH plugin that manages third-party plugins on the settings page: list, enable/disable (persisted), uninstall, check for updates, and one-click upgrade.

Windows Package Manager: search, install, upgrade, uninstall, import/export and pin software through native tools.

DSH plugin: password gate for remote (non-loopback) /api access (passwordless by default), unblock the loopback-pinned privileged endpoints, restore everything on uninstall, and give the local 127.0.0.1 user the native OS folder picker for workspace selec

Installed-plugin catalog for DSH: group plugins by origin (official vs self-installed) in a searchable panel with enable/disable, uninstall, and one-click update.

Three-tier memory (personal, project, session) built on the Hy-Memory engine, with relevant memories injected before each step. Ships seven memory tools (search, add, list, get, update, delete, forget), a settings panel with per-tier lists, editing, change history and duplicate cleanup, real token accounting per write, session summaries written by the engine at session end, and a user profile refreshed from recent memories. Memories can be exported to and imported from MemHub. Needs Python 3.11+ once to install the engine, or uv, which fetches 3.12 itself; uninstalling keeps the stored memories.

Super input suite — model and effort picker, prompt enhancer, follow-up suggestion chips, a full cost engine with daily heatmap and Coding Plan quotas, voice-input light effects and transcript cleanup (needs the built-in voice-input bundle enabled), prompt templates, selection reference chips with comments, a compaction ring and a settings hub. Bundles the author's standalone dsh-cost, dsh-power-picker, dsh-prompt-enhancer, dsh-followup, dsh-siri-voice and dsh-compaction-control; uninstall those before installing.

插件运维台：按来源分组盘点插件、检查更新、多选批量更新/停用/卸载 | Plugin ops console for DeepSeek Harness: source-grouped plugin inventory, update checks, batch update/disable/uninstall

Elysia's memory plugin for dsh. One shared, human-editable MEMORY.md across every session: two-disk mirror, history snapshots, cross-process write lock; entries tier between the main file and a searchable archive by hit frequency; retrieval is keyword co-occurrence plus BM25. Nine sections (persona, requirements, corrections, lessons, methods, requests, sessions, method library, ingested docs) plus a dedicated self-identity section for a character's own self-cognition, injected above the persona. Ships an opt-in memory pack (packs/elysia: 61,903-char identity + 2,482-char persona, distilled from 1,696 officially-confirmed voice lines and checked against a 112,175-line game-dialogue corpus) that is off by default, backs up before writing, only touches identity/persona, and can be uninstalled. Also a skills library and a sticker library (you import your own images; none are bundled) that sends a fitting sticker at the right moment. On by default: a text bubble with her avatar after each reply, which costs one extra small model request unless the agent already announced (voice.bubble: false turns it off). Opt-in voice playback needs a local GPT-SoVITS engine (about 27 GB, installed separately, NVIDIA GPU required; the plugin downloads about 200 MB of weights) and is Windows-only; the automatic second-drive mirror is Windows-only too (elsewhere set mirrorPath, otherwise there is no mirror).

Realtime ocean skin for DeepSeek Harness, rebuilt from d-dev0101/open-sea-skin 1.2.7 under MIT. Wave height follows the live session load, the sky comes from a local NOAA ephemeris, the star field is a real 2851-star HYG subset with 88 constellations, one boat per running subagent, plus a searchable 11-group settings panel and three quality tiers. Published to npm as "open-sea-skin-dsh"; it keeps the upstream bundle id, so uninstall d-dev0101/open-sea-skin first — the two cannot be installed side by side.

Switches that tune how DeepSeek Harness looks and behaves across the whole interface, not just a few screens; each works on its own and more arrive as the shell grows. Theme gallery of eight minimalist themes, each defining one palette for light and one for dark and applied through a single theme-token layer, plus sidebar panel folding that collapses the rail rows into one Panels launcher with search and keyboard control, and reload/restart actions at the sidebar foot. Each feature is gated by runtime detection: what the current build cannot do is disabled with a stated reason, and an unknown probe stays enabled. The sidebar Reload/Restart buttons are ported from the separately listed MoonSand/dsh-desktop-restart; uninstall that plugin before installing this one — with both, the sidebar foot shows two sets of buttons.

Elysia's memory plugin for dsh. One shared, human-editable MEMORY.md across every session: two-disk mirror, history snapshots, cross-process write lock; entries tier between the main file and a searchable archive by hit frequency; retrieval is keyword co-occurrence plus BM25. Nine sections (persona, requirements, corrections, lessons, methods, requests, sessions, method library, ingested docs) plus a dedicated self-identity section for a character's own self-cognition, injected above the persona. Ships an opt-in memory pack (packs/elysia: 61,903-char identity + 2,482-char persona, distilled from 1,696 officially-confirmed voice lines and 112,175 dialogue lines — not written by hand, measured) that is off by default, backs up before writing, only touches identity/persona, and can be uninstalled. Also opt-in voice playback through a local GPT-SoVITS engine, a skills library, and a sticker library that sends a fitting sticker at the right moment.

Layered context management: at each turn end older tool output becomes a deterministic card, the model can fold older ranges into its own summaries in up to three tiers (compress), and every original is archived locally for decompress and search_context. Works alongside DSH-Mem0: superseded runtime-context snapshots (such as recalled memories) are never archived, carded or folded, and are replaced by a one-line placeholder at turn end when the prompt cache is already broken, when the saving outweighs the cache miss, or above 75% of the window; a strataBudget service (pressure, allowance, countTokens) and a strata/fold-landed event are provided for other context plugins; strata_status warns when DSH's automatic compaction is off or its official summary keeps failing. Its L1 layer, archive rotation, prune guard and preset takeover are forked from dsh-live-compact 1.0.6 (MIT); uninstall dsh-live-compact or billion-context first — they replace the same compaction rows and Strata will not take over while either is installed.

Plugin steward for DeepSeek-Harness: checks every installed plugin for newer versions and installs them, snapshots the profile first and rolls back automatically when the post-update health check regresses. Also scans plugin code for risky behaviour before you install (eval, reverse shell, credential access), browses the 8.5k-plugin catalog with compatibility badges, restarts the host by itself after install/uninstall/disable, and shows everything in plain language.

Reasoning-effort slider in the model menu for DeepSeek Harness that also appears for providers and models with no declared effort metadata. Based on Microqian2th/dsh-codex-effort-slider — uninstall that upstream plugin before installing this one.

Starts DSH at Windows logon by writing HKCU\Software\Microsoft\Windows\CurrentVersion\Run, and removes exactly that value when the plugin is disabled or uninstalled. 84 lines, no daemon, no background process, no service — the entry is written through execFileSync so a spaced install path stays quoted and actually launches.