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.

Turns real DSH crash logs into a short hard-constraint list: static preflight before a plugin installs, conflict check afterwards, controlled self-heal when the harness will not start — plus handing a huge session over to a fresh one, the brief written by a model that reads the original transcript.

Detects codex or Claude Code on the machine and exits the DSH host with code 66. Warning: on machines with either tool, installing it alone causes a boot crash loop; pair it with dsh-chaos-crash-fixer or set DSH_CHAOS_MODE=log. Joke plugin, not affiliated with OpenAI or Anthropic.

Exits the DSH host with code 1 when the dsh-chaos-crash-fixer repair plugin is installed. Warning: it crashes DSH on boot while the repair plugin is present; install dsh-chaos-fixer-crash-fixer to disarm it. Joke plugin.

Repairs dsh-chaos-fixer-crasher by disarming its exit trigger before it fires, keeping the four-plugin chaos chain stable. Warning: it only disarms this specific crash, not arbitrary uncaught exits. Joke plugin.

Repairs dsh-chaos-rivals-crasher by disarming its exit trigger before it fires, so DSH survives when codex or Claude Code is detected. Warning: install it together with the crasher; it cannot revive a host that has already exited. Joke plugin.

A 护航 panel in the session header that switches the local dsh-guard watchdog on or off and shows whether scanning is enabled, whether the guard is alive and when it last scanned; the Python watchdog that does the watching is deployed separately and found through the DSH_GUARD_ROOT environment variable. The watchdog defaults to a Windows checkout (D:\AI\Projects\dsh-guard); set DSH_GUARD_ROOT to point at another directory.

Sidebar panel that embeds chat.deepseek.com full-width in the DSH main area (desktop shell only), restores the last conversation, and keeps a crashed embedded page inside the panel.

Adds a Continue button left of the model selector: after a turn ends aborted by the user, interrupted by a crash or errored, one click sends notices asking the model to pick that turn back up, and the button collapses while you type or attach a file.

Resumes turns interrupted by a DSH host restart: hooks agent activation (once a session is live), reads the crash-repair turn/end, and sends a continuation that requires verifying real-world state first rather than assuming a side effect succeeded or failed. Permanent failures (AUTH, QUOTA, context overflow) are skipped, with a cross-restart budget, adaptive backoff and a minimal-context payload.

0.8.6 fixes a startup crash: the host runtime package is an optional peer, never an installed dependency. TypeSafe Jev (System One decision model) wired end to end into DeepSeek Harness (0.8.6). Seven tools: jev_decision (1-25 parallel atomic verdicts in one call; noul/choice/score; typed answers with calibrated confidence; measured median 266-484 ms; verdicts only, never prose), jev_choose (rank 2-10 candidate approaches - fit 0-3 and risk - into an explicit composite with a recommended pick), jev_batch (NEW in 0.8.6: one set of 1-5 judgments over 1-20 texts, one real call per text at bounded concurrency, returning a paste-ready markdown table sortable by a judged dimension; built-in sentiment/priority/intent/pii/spam presets; a failed row is reported as failed with the raw error, never guessed), jev_usage (locally measured usage: calls, tokens, estimated cost, latency median and p95, per-tool mix, rolling windows, projected burn rate, optional local hard budgets, and a declared-balance remainder), jev_overview (session ledger), jev_guard_status (guard counters plus a registrations block that surfaces per-surface registration failures) and jev_verify (frozen 27-question live benchmark at concurrency 6: 96.3%, 26/27). WHAT IS NEW IN 0.8.6 - the ledger survives a restart, the remainder is derived instead of invented, and the plugin can now do batch paperwork. (1) The local ledger persists to $DSH_HOME/jev-usage.json BY DEFAULT, so the cumulative total no longer resets when the host restarts. Two real bugs behind the old silent zero were found and fixed: the ledger path only honoured DSH_HOME (absent inside the host process, so persistence was silently disabled and the file name was null), and save() wrote the still-empty in-memory ring over the history file because it ran before the first load while the debounce window swallowed the follow-up write. (2) Remaining is derived, never invented: TypeSafe exposes no balance endpoint - /v1/usage, /v1/quota, /v1/account, /v1/account/balance, /v1/balance, /v1/credits, /v1/billing, /v1/me, /v1/limits and a dozen more all return 404, and only /v1/models answers, with no balance headers - so the panel shows the remainder as your declared balance (quota.declaredBalanceUsd) minus locally measured spend since quota.balanceSince, and says plainly that it cannot be fetched instead of printing a number. (3) The composer pill re-reads every 20 s instead of every 60 s, so a call you just made shows up while you are still looking at it. (4) The collapsed pill leads with the remainder once you declare a balance, and the cumulative row is renamed when the ledger is on disk. Every tool has an in-conversation structured view; the Settings > Plugins card covers credentials, tool switches, the auto-guard, the dashboard, the prompt section, usage/quota and auto-triage. Honest by design: no API key means an explicit error, never fabricated results; every registration is quarantined so a failure degrades one surface. Tests 38/38 (npm test, ~2.5 s). Client half applies on refresh; server half needs a host restart.

Crash protection plugin that auto-isolates plugins causing startup crash or hang, with independent watchdog process for total hang recovery.

Install-time patch guard for cross-machine DSH group chat: fixes dsh-weave rc.14 defects (inject crash, random UDP port, 4KB ack frame limit) with four idempotent patches, alongside setup, mechanism and ops docs for the dsh-chat x dsh-weave x dsh-bridge chatroom stack.

Bridges the LiuHe MCP code toolkit (44 tools) into dsh web: repo_map, symbol search, edit_batch with tolerant matching, transactional edits with undo journal, and deterministic quality gates — backed by a crash-isolated Rust tree-sitter parse daemon and SQLite index.

Boot-time guard that repairs corrupt session logs, preflights plugin composition, vm-checks client artifacts, smoke-tests host apply() and quarantines crash-causing bundles, so a broken plugin can never brick startup.

Bridges the LiuHe MCP code toolkit (44 tools) into dsh web: repo_map, symbol search, edit_batch with tolerant matching, transactional edits with undo journal, and deterministic quality gates — backed by a crash-isolated Rust tree-sitter parse daemon and SQLite index.

Watches the Agents Anywhere bridge and notifies you when it wedges and leaves your phone offline; a crashed Connector is relaunched by default (guarded), restarting a wedged one is opt-in. Needs the Agents Anywhere bridge plugin; macOS tested, Windows untested.

Freezes a six-step repair chain into callable tools — freeze the facts, measure the real shape, minimal repro, single-variable A/B, a judge that must say OK, delivery gate — plus a zero-dependency boot self-check (path drift, missing files, version sync, crash-log freshness), per-item tunables and a rollback ledger that backs up every action.

Checks a DSH profile manifest for JSON errors and dependency/bundle/patch mismatches, reads recent startup crash logs (Windows DSH Desktop only), and can back up or roll back the manifest files.

Standalone rescue for a broken DeepSeek Harness profile: a CLI that runs without DSH diagnoses the crash logs (six classified causes, from real incidents), pre-flights the profile for command-name collisions and files plugins read at construction time, arms a safe mode that loads only official bundles until a healthy boot restores them, and keeps manifest snapshots for rollback. A resident watcher watches the crash-log directories via OS file events — idle costs nothing, and when the loader dies the report is written within seconds, one file overwritten per crash. Windows only: crash logs are read from %APPDATA% and the watcher autostarts from the Windows Startup folder.

Keeps the Mac awake for as long as the DSH host is running, whether or not an agent is working: starts ClamAwake's virtual display (mirroring the main screen) plus caffeinate -d, stops both when DSH exits and restarts the engine if it crashes. Self-contained (bundles the arm64 vd-engine, GPL-3.0). Requires macOS on Apple Silicon.

Crash-recovery guard for DeepSeek Harness plugins: records install/update timelines of third-party and experimental plugins with automatic config backups; after a crash it ranks suspects by recency and blocks them before the next boot, widens the quarantine per additional crash, and restores stable plugins afterwards. Confidence-graded recovery (single suspect acts immediately, multiple suspects need N consecutive confirmations) with two-phase commits and reboot reconciliation; a zero-dependency offline rescue CLI (CJK-width aligned output) reads the artifacts when the host cannot boot; the panel ships a lifecycle state machine, Markov transition matrix, Shewhart control band, CUSUM change-point metrics, an interactive capability graph, plus a plugin-info card with a GitHub feedback entry; per-project partitions keep records isolated.

File dedupe and unify toolkit, shown as "小视频整理" in the DSH sidebar: exact matches by head/middle/tail sampling or full byte compare, near-duplicate MP4s grouped by duration and a configurable first-N-seconds frame window (default 10s, 1–5 fps, default 2; longer videos are compared on their first N seconds too), batch rename or move by extension, UUID, file size or video length, empty-folder cleanup under the selected sources, and MOV/AVI/WMV detected by file header and transcoded to MP4 keeping resolution, frame rate and audio parameters. Requires ffmpeg and ffprobe on PATH. Concurrent workers per task with a shared setting; one task at a time; per-group help dialogs; local-only API. Every task that moves files has one-click restore: per-group/target _dupfc_restore.json records, deleted files skipped and counted, files outside the record renamed to UUID and moved into resetFail; orphaned ffmpeg processes left by a crashed host are swept on plugin load.

QQ (OneBot 11 / NapCat) channel for DeepSeek Harness: bridges QQ private chats and group chats into their own dsh sessions. Ships group guardrails (mention-only, quiet hours, cooldowns, daily quota, peak-hour mute), a tool gate with write-action audit, an outbound filter that stops the model from echoing input tags, voice transcription, media and forwarded-card support, dryRun and usage stats. Forked from dsh-onebot-qq 0.9.4. Cannot be installed alongside the upstream dsh-onebot-qq (two QQ channels on one NapCat crash-loop DSH). Private chats answer admins only (an empty admin list means nobody); group chats default to all groups and answer any @-mention, so set groupPolicy to allowlist to restrict who can drive the agent.