Plugins

Composer-side prompt library: a button next to the input opens a picker with group chips and word-split AND search over saved snippets, inserting one into the draft or appending it, filling {{variables}} before insertion, with 13 built-in templates. Snippets live in one local JSON file behind a rev lock, a per-save backup, and whole-library JSON import/export; a settings page handles editing, grouping, pinning and a 6-second delete undo.

Adds an Optimize prompt control next to the composer send button that rewrites the draft into a structured prompt (plain, sections, or role-task-goal) streamed from the session model, with Use, Undo, Copy and Close.

Rewind conversation turns and checkpointed file changes while restoring tasks and evidence at the same history boundary.

Bridge to the UnrealAgentLink editor plugin: the Unreal Engine editor dials this WebSocket bridge on 127.0.0.1:17860 and the agent drives it through 12 tools covering all 170 cataloged commands (actors, levels, blueprints, materials, PCG, screenshots, logs, undo).

Markdown visual enhancement and session workbench for DSH Web Composer: live syntax presentation with caret-row source reveal (inactive-line markers can be hidden or faded), deterministic local completions, and list/quote continuation on line break with ordered-list renumbering; long pastes fold into persistent chips with inline context or file-mode delivery (@path across 50+ extensions) plus sidebar LLM rewrites (stop, preview diff, undo/redo); real-time context map with token gauge, composition breakdown, per-turn growth chart with anchor lines, in-place /compact, turn forking, and a Context Doctor workspace (overview, diagnostics, analysis, feedback).

Session cleanup for the DSH sidebar: drag a conversation row onto the balance widget's whale and it eats the row — the session directory goes to a recycle bin, not to rm. Deletion asks one short question first (how many tokens it burned, and that prompt can be turned off in Settings). Undo from the persistent toast or the 12-entry log in Settings. Clicking the whale jumps straight to the plugin's Settings, where you can swap the avatar, drag the mouth into place and rewrite all 17 UI strings; blank, currently-open and running sessions are refused in place with a reason.

Hide chosen workspaces from the DSH sidebar and archive their sessions along with them, so those sessions never spill into the ungrouped list. Fully local and reversible; with dsh-archive-manager installed the archiving is real and undoable, otherwise it stays a display-only layer. Restore any hidden workspace from Settings.

Transactional uninstall engine: every destructive action runs validate/preview/execute/undo with Saga rollback, WAL crash recovery, hash-chain audit, hardlink dedup, and a Bayesian oracle that predicts success probability before you commit.

Rollback controls for the latest completed conversation message

Session-scoped Markdown writing pad docked beside the conversation, with edit/preview modes, selection-based AI rewrites, diff review, undo, and draft recovery.

Read-only judgment layer for DSH profiles: explains row by row why a plugin was refused admission or silently disappeared, pre-flights a runtime upgrade against the installed set in four buckets (would-be-denied, cannot-tell, improved, likely-fine), and can measure a candidate in an isolated DSH_HOME. Defaults to read-only; the two --apply verbs only write compatibility.json or dsh.profile.bundles, each after a backup, with a journal and a byte-exact --undo.

Ordered and unordered list editing in the DSH composer: automatic numbering and renumbering, marker-aware caret moves, level promotion, and an undoable AI prompt-polish button.

Running panel for dsh for skill management and visualizing skill calls: live skill and tool calls, the skill-catalog token budget against a line that follows the model's context window, a skill loadout window that enables and disables skills in bulk with dry run and undo, goals and ideas, and saved conversation excerpts. Applying a loadout moves skill folders between ~/.agents/skills and ~/.agents/skills-parked, which other apps share (dry run first, undo available); automatic translation, goal inference and idea collection call your default model only after you opt in on first open.

Physically delete DeepSeek Harness Sessions with an undoable recycle bin: move a Session log out of its workspace, restore it, or purge it for good.

Per-turn, per-file checkpoints for files written by DSH tools: keep, undo, redo, and revert to here from a dock above the composer, with inline before/after diffs, pinning, and per-session storage cleanup.

Conversation rewind for DSH: rewind in place to any earlier user message with one click or the /rewind and /undo commands, without a new branch or window, and optionally restore workspace files edited by write tools. A Settings page lists recent rewinds per workspace and can undo the file changes from the latest one. Derived from SiriLee/dsh-rewind and maintained separately; do not install both (they share the /rewind and /undo commands and the rewind-snapshots directory).

Moves a workspace to another folder from its right-click menu: renames the folder, carries its nested workspaces and sessions along, rewrites their records, rolls back if anything fails, and can be undone in one click. Developed on Windows; macOS and Linux share the same code path and are exercised in CI against a real harness host.

Workspace-wide conversation tree for DeepSeek Harness: every conversation and branch on one canvas with its source turn, own token usage and notes; compare two branches, bring a summary back, export a Markdown outline, archive (undoable through DSH's native archive), and delete a whole subtree (macOS and Windows only; deletion runs immediately and is permanent).

Recycle bin for conversations - delete a session from the sidebar into a trash folder, undo within 8 seconds, restore or delete forever later.

Prompt booster for the composer. Left-click the composer button and the draft is lightly polished and replaced in place (undoable, no popup). Right-click opens a menu with 8 built-in modes plus your own, where each mode's template can be edited or new modes added (saved on the host), and the model used for the rewrite is picked from the live dsh model list. Other modes show a before/after preview. Based on rongxingda/dsh-prompt-enhance (Apache-2.0); do not install both — both add a composer button and share Ctrl+Alt+E, so disable one of them if the upstream is present.

A project kanban in the dsh sidebar. Every project gets its own section with four fixed columns (todo, in progress, feedback, done); add, rename, recolor, reorder and delete projects and tasks, advance a task with one click, reorder by drag and drop or with the card menu, undo a deletion, and collapse a project. Clicking a project opens a page for its key data and results: two trees (data on the left, results on the right) you arrange by drag and drop and merge into chapters, lines that show which result came from which data, stars with a star-only view of the same tree, relations between results, colours by category, and undo or redo with Cmd+Z. Colours, radii and shadows come from the dsh theme, so it follows light and dark, and the layout reflows from four columns to two to one.

Disk cleanup for Mac (macOS) and Windows: a risk-tier audit and space breakdown, then moves caches, leftovers and similar photos to the Trash or a quarantine folder, with undo. Space is freed only after a batch is permanently deleted.

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.

Agent enhancements for DeepSeek Harness Desktop: prompt history, one-click optimize with undo, next-prompt prediction, conversation and tracked-file restore, model bill, context details with compact, Auto routing, and per-provider DeepSeek balance.