Plugins
Browse, filter, and install DeepSeek-Harness plugins.
315 plugins found
dsh-textbook-parser
teanuky/dsh-textbook-parser
Breaks a STEM textbook into a four-level card map (book, chapter, section, knowledge point) in a sidebar panel, where each point names the defect of the previous approach it fixes and each chapter keeps its remaining gap as the motive for the next; marks key points with a star, shows theorems as cards with statement, conditions and symbol table, folds step-by-step derivations, and exports the same map as a self-contained offline HTML file, Word or PDF. Ships a textbook-parser skill and a textbook_archive tool (save_book / save_chapter / list_books / get_book / delete_book) plus a bundled 12-chapter Signals and Systems sample with 116 sections, 263 points and 601 worked examples. No runtime dependencies and no network access.
dsh-payload
shengpj/dsh-payload
Adds a { } button under each assistant reply in the DSH Web GUI that opens the exact request sent to the model for that reply: model, call parameters, messages, tools, and provider-reported token usage.
dsh-eni (armor-break)
konshinhaoshin/dsh-eni
Per-model jailbreak armor: a Settings table arms model patterns with spells (25 spells: 20 across 11 model families, plus a model-agnostic armor, an upstream preferences file and three stackable persona layers), previews the exact injected block, and writes it into the armed model's system prompt only, leaving every other model untouched.
dsh-local-trim
leonlee114514/dsh-local-trim
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-calculator
qqiqiq/dsh-calculator
Adds a calculate tool backed by a hand-written parser with no eval, batch evaluation and an exact-decimal mode for money math.
yueling
sdsaewq/yueling
Yueling (月令) — a solar-term card for the DSH Web UI: a dot under the conversation header unfolds into today's term, its three pentads, how long until the next term, a day-pillar reading, and a year table of all 24 terms. The hexagram page casts coins and keeps a seven-day log; the jieqi skill is bundled.
dsh-prompt-optimizor
grampusworld/dsh-prompt-optimizor
Rewrites the composer draft into a tighter prompt in one click and restores the exact original in another, running on Harness's own models by default or any OpenAI-compatible endpoint, with an editable polishing prompt.
dsh-project-groups
thinkofrain1213/dsh-project-groups
DSH plugin: group sidebar sessions by project (a label, not a directory), give each project a work document whose format follows a spec, register its related links, and show the exact runtime-context injection on the settings card.
dsh-ui-overrides
cholot-zhou/dsh-ui-overrides
File-based CSS overrides for the DeepSeek Harness Web UI — edit ~/.dsh/ui-overrides.css and reload to restyle any plugin's elements, for example making the quote bar fully opaque.
dsh-pouch (dsh-hooks)
moon16u/dsh-pouch
File-defined command hooks across the session, prompt, tool, permission, Stop and PostCompact seams, with an approval-only GUI: no unapproved command runs, and approval binds the exact definition, canonical working directory and pinned watchFiles. Commands run as the DSH OS user without a hook sandbox, so the trust ledger is explicit.
dsh-simple-tps
flyxiu/dsh-simple-tps
Per-turn first-token latency and output speed beside the built-in usage pill, folded on the host over the durable event log: first-token time is replayed from each message's compact stream, so historical turns stay exact after a reload rather than showing an estimate.
dsh-poker-night
dingchenhui0618-arch/dsh-poker-night
A no-limit Texas holdem table where you are the only human. Eight seats: you plus seven LLM agents written on AI-founder personas (Altman, Dario Amodei, Musk, Liang Wenfeng, Jensen Huang, Lei Jun, Zhang Yiming). Each agent bets through the currently selected host model and falls back to its own heuristic style when no model is available; every agent only ever sees its own hole cards, because the prompt is built from that seat alone. A hand ends by revealing the winner hole cards, with a flip animation replayed on every community card, and a rail keeps a live action-and-table-talk timeline. LAN mode (off by default, single-machine play is unaffected) lets another install on the same network join this very table: the host shows a six-digit room code, UDP broadcast discovers nearby tables, and a joiner randomly takes over one bot seat, with the seat handed back to its character on leave or after twenty seconds of silence. The room code is never put on the wire and every snapshot is generated for one viewer seat, so nobody, not even the host, can read another player hole cards. Fork of BubblePtr/dsh-holdem (MIT), credited in the README and in the UI.
dsh-plugin-skill-autoroute
soh4c4759/dsh-plugin-skill-autoroute
Run one deterministic skill-routing pass on every user instruction, so the skill catalog is consulted before the model answers instead of after it remembers. On the host it scores the live catalog from the instruction's keywords and the catalog metadata, with no LLM or embedding call, and appends a single producer-tagged message to the step: a ranked shortlist by default, the winning SKILL.md body once the score clears a threshold, or an instruction to call a named router skill exactly once. Unrelated instructions stay silent.
dsh-skill-intelligence
polinnizhong/dsh-skill-intelligence
Read-only Skill inspector for DeepSeek Harness. It shows which Skills a conversation actually loaded — from observed loading evidence only — and which Skills the current environment can discover, and gives each Skill a detail page whose left column switches between the framework, the evidence, validation, evaluation, step evidence, run logic, the before/after of a modification when there is one, the definition facts and the raw SKILL.md. Declaration and run stay apart: the body is split into structure, declared workflow and progressive disclosure, the run logic is read from local receipts, and every declared step carries its own evidence — no observed evidence is reported as exactly that, not as executed or not executed, and the run's instruction hash sits beside the current definition as match, mismatch or unavailable. The evidence view gives each subject one of three states (declared, observed, unavailable), walks definition → load → use → outcome, names what each stage cannot reach, declines to attribute an outcome to the Skill, and can copy the snapshot out as byte-stable JSON. Validation checks the Agent Skills open standard and the platform profiles separately, each finding on a rule id and labelled standard compliance or platform compatibility. Evaluation keeps repeatable cases with the conditions each run was recorded under and the four stages of runtime evidence, comparing two runs of one case item by item without scoring them. The trigger boundary keeps should-trigger and should-not-trigger prompts, has you run them in a real new session, then records what was observed — whether a load record appeared and whether it was model-invoked or user-explicit; a typed /name is not a natural trigger, and an absent load record is reported as not observed, not as not triggered. The same projections are exposed to the agent as a read-only skill_review tool, which returns a structured review packet and says what it omitted when the packet will not fit whole. The inspector, the evidence chain and these reviews are computed on this machine from the definition and the session receipts; the Chinese reading version sends the body through the host's model when you ask for it, keeping code fences, paths and frontmatter verbatim and storing the result on this machine. A Skill can also be cloned into a local copy. It does not write back to the source Skill and does not run its scripts.
dsh-mol (dsh-ui)
yijian-quiet/dsh-mol
A chemistry workbench panel in the DSH web UI. Draw structures on a Ketcher canvas (a script downloads the build; it renders locally in the browser with no backend), then compute on the current molecule with RDKit without calling a model — formula, molecular weight, exact mass, logP, TPSA, hydrogen-bond donors and acceptors, ring counts, Fsp3, stereocentres, Lipinski and Veber rule-by-rule checks, QED, PAINS/BRENK structural alerts and the Murcko scaffold, plus SMARTS substructure matching, fingerprint similarity, format conversion, standardisation and batch clean-up. A retrosynthesis tab talks to a pluggable backend (Retro* by default) over a single JSON-in/JSON-out process, and states what is missing when the backend is not installed instead of inventing a route. The calculations run in this repository's Python package (rdkit), which is installed with pip separately.
dsh-action-outbox
jimchen-g/dsh-action-outbox
Stages exact DSH tool calls without dispatch, presents complete canonical arguments in a durable Batch Review Inbox, invalidates approvals after edits or restarts, and commits only a matching SHA-256 digest plus single-use nonce through the normal DSH tool pipeline.
dsh-approval-explain
martlet-tech/dsh-approval-explain
Adds an Explain button to the approval card: one model call returns exactly three lines reading the pending operation, covering what it does, which files or system state it reads, writes, modifies or executes, and a risky/attention/safe verdict with a one-sentence justification. It also renders the detail area for `write` and `edit` calls, which the shipped renderer leaves blank because it only reads a `command` field.
dsh-word-vault
wangzhanchao883/dsh-word-vault
English word bank for one or more children, built for DSH 0.2 and newer (0.2.0-rc.1 and 0.2.0-rc.2 verified, including the desktop shell; 0.1.7 through 0.2.x all work). 1.10 added pronunciation: in the memory cards and in the online quiz every word and every example sentence has a speaker button that reads it aloud with the built-in Windows voice - offline, with no network call and no model call - and the whole batch of audio is pre-rendered in the background whenever cards or an exam are generated, so a click starts instantly. 1.11 adds: pick words in bulk straight from the library (10 at a time, top-down, up to 50 selected, with a live counter); turn the words a child got wrong in a quiz into a one-click review set - an online card page that speaks plus a printable PDF; and every card example sentence now carries its Chinese translation underneath. Words marked on textbook or workbook photos (highlighter, red pen) are found by their marks, then translated and stored in a local SQLite bank with an occurrence count; the clipboard picker, the photo route and dictation all land in the same bank. High-frequency words surface by count, three correct answers in a row flag a word as learned, and any selection prints as a memory card set (a word split into chunks, real IPA per chunk, a grade-7 example sentence, a mnemonic hook, an A4 sheet of eight) in HTML, PDF or Word, with a print-and-answer exam page whose four options are set against each other for meaning. On 0.1.5 the settings panel does not appear - install dsh-word-vault@1.0.2 instead.
dsh-github-mcp-hint
ziye1208/dsh-github-mcp-hint
GitHub MCP helper panel: shows copyable example prompts at Settings → Plugins → GitHub MCP (random 4 from a 30-example pool), visualizes a public user's repo star/fork/language, and adds the gh_repo_stats and gh_current_user model tools.
context-razor
weibaohui/context-razor
Context trimmer: lists the current session's context entries with role, preview and ≈token estimate (cl100k), highlights over-threshold items, and removes selected entries exactly without LLM summarization.
dsh-split-screen
yeyeyeyeshifu/dsh-split-screen
Split-screen workspace for the DeepSeek Harness web GUI - chat with several agents side by side in panes that look and behave exactly like the stock conversation: drag-to-split, live sessions with queued messages and in-pane ask_user/approval cards, per-pane view tabs (Chat/Trajectory/plugin views), a compact TUI composer mode (Alt+T) and pane-focus shortcuts. Requires DSH 0.1.5+.
dsh-token-speed
gbeta/dsh-token-speed
Draggable ring gauge showing live model output speed (tok/s) with an expandable per-step detail panel (exact speed, TTFT, cache-hit rate, cache read/write, reasoning and input tokens, model, cumulative tokens).
dao-zang-skill
godners-code/dao-zang-skill
Daoist Canon (DaoZang) retrieval skill: keyword or semantic search over 285,117 scripture chunks with exact original-text extraction (line numbers, hit markers), and one-command workspace setup from the Godners/DaoZang dataset.
dsh-conversation-timeline
bycall/dsh-conversation-timeline
Conversation timeline dock bar for DeepSeek Harness: a turn-level timeline above the composer with user-prompt hover tooltips and click-to-jump back to the exact chat position.