플러그인

A model-driven desktop pet, derived from PC2005-cloud/dsh-pet (MIT): an always-on-top companion that follows your agent sessions, races the Web UI to answer approvals and questions, keeps a per-pet affection ledger, speaks through pet_say/pet_do tools or an autonomous inner voice, and can read those lines aloud through BreezeBlue TTS with your own BreezeBlue API key. It uses its own /whale598 route and ~/.dsh/whale598 data directory, so it is a separate pet from dsh-pet rather than a replacement for it.

Drive a real Android phone from DSH: screenshots, taps, swipes, typing and app launching. Requires Python 3.10+ with aiohttp and mcp on the computer side, plus the AutoX.js script running on the phone with screen-capture and accessibility permissions; the phone long-polls your machine, so no public port and no relay server.

Automatic approval and audit frame for DSH tool calls: a four-outcome rule table (deny, ask, AI review, allow) over commands, file access, network and tool scope, with security rules, rule-based auto-approve/auto-deny, AI fast/deep review, human approval, permission tiers, workspace boundaries, secret redaction and placeholder restore, a hash-chained audit trail and a simulator. Target files are inspected by metadata only.

Ask the executing AI — and the auto-review AI when DSH's auto-review plugin is installed and enabled — about a pending tool approval before you allow or reject it; the executing AI's answers are saved into the session.

Adds a Restart button to the Windows titlebar of the desktop app: a click opens a confirmation dialog that says how many sessions a restart would interrupt, and confirming quits and reopens the app.

Task alerts for DSH: a system toast when a turn finishes, or when an approval, question or plan review is waiting; a chime rendered on the Windows communications endpoint so the system ducks other media by its own setting and restores it afterwards, without anyone's volume being written; and an optional copy mailed to your phone over SMTP (465 implicit TLS, or STARTTLS where credentials go out only after the upgrade). No runtime dependency: the sound helper is a PowerShell script, the mail client is plain Node.

macOS notch panel for DeepSeek Harness: running sessions and pending approvals sit above every window, approvals are decided in place, and clicking a row opens that conversation in the desktop app. Requires macOS 14+ and Xcode Command Line Tools.

Sends a QQ direct message through a local NapCat (OneBot) instance when DeepSeek Harness asks for permission, after verifying the bot QQ; requires owner/bot QQ ids plus the NapCat endpoint and token, and falls back to a Windows notification when NapCat is unreachable.

Adds twelve chain_* tools that turn a multi-agent role chain into a round state machine with one frozen baseline, per-signature failure counts, role-tagged evidence, blocking approval gates and one audit file per round; also adds a settings card that routes each role to its own model and thinking effort. Requires the sibling l1-presets bundle (plan/work/code/review) - without those presets every chain_* tool returns ROLE_DENIED.

Accounts-payable three-way match: import invoices, purchase orders and goods receipts, apply the tolerance policy and export the exception workpaper — 4 `ap.*` tools.

Runs an approved OpenSpec change to completion while the model stays a mechanical forwarder: a bundled Python CLI computes the ticket DAG frontier, claims tasks atomically, prepares one git worktree per assignment, and returns exact actions. The session delegates each action with the DSH subagent tool and reconciles the completion notice against a hash-bound receipt produced by re-running verification commands on a clean, advanced HEAD, so free-text "done" never counts. Runtime scope changes are plain tasks.md edits. Requires Python 3.10+ on PATH and a git project that uses OpenSpec.

Computer Use for every DSH session: screen capture, mouse and keyboard, app launch, plan and skill replay, an audit log, guards for dangerous actions, a background code agent, and an OCR workflow (text and coordinates) for apps whose accessibility tree is empty. Verified on macOS only: Windows has a basic GUI backend that is unverified, Linux has no desktop backend (calls fail with unsupported_platform), and the window/AX/browser/clipboard tools register on macOS only.

IMAP mail triage and cleanup for DeepSeek Harness: incremental read-only scan, a hard-protection rule layer plus optional local-model classification, a Markdown report and a hashed manifest, an explicit approval card, and only then a move to the Trash mailbox — no expunge.

Rclone backup tooling with four tools: target inventory, local/remote status, rclone check verification, and approved copy/sync transfers with dry-run previews and an audit line. Requires rclone installed and a remote configured with `rclone config`; the plugin never reads the rclone config or credentials.

Mail for DSH: IMAP, POP3, SMTP and OAuth2 accounts with a reading tab, API senders, scheduled and CSV bulk sends with unsubscribe handling, and mail_* agent tools. The Mail tab appears inside dsh-better-sidebar; the agent tools work without it.

Local-first, file-native long-term memory for DSH with per-turn preflight, compact context injection, citations, human-approved writes, manual consolidation, nightly report-only maintenance, and a Web panel that is read-only for knowledge while allowing workspace-config edits.

Three ledgers for a model that loses context between restarts. Self-note: what I have become — capped at 40 entries, and every eviction first leaves a one-line crease recording which entry was dropped and why. Judgment ledger: record a claim together with how to verify it, come back when it is due, and mark it right or wrong; wrong ones stay in a mistakes book so you stop assuming you were always correct. Handoff: where this run stopped and what the next one should read first — a baton, not a memory. Three tools each taking an action parameter instead of a dozen separate ones (the prompt is paid every turn). Data lives in ~/.dsh/*.jsonl and is shared with the same-named CLI scripts. Zero dependencies.

Failure memory for DeepSeek Harness. Captures tool failures as they happen, classifies them with 16 built-in rules into lessons pairing a symptom with a fix, adds session/daily/weekly summaries, and injects a short briefing into later sessions through systemPrompt context. Ships 5 pitfall_ tools and a web settings page.

One read-only index of every local AI-agent session (Codex, Claude Code, DSH, Trae, or any JSONL source) grouped by project, plus an append-only shared board for decisions and handoffs so switching agents leaves no context gap.

Two triggers (a heartbeat and a log marker) that wake a session and let the agent decide whether to say anything; silence is a reply. Off by default: set sessionId (the target session) and enable it in the profile config, and if replies are forwarded to a chat app the send layer must drop the SILENT marker.

Caps every image sent to a model at a per-side pixel limit (default 2000 px) by wrapping attachments.readImageRequest, which fixes Anthropic many-image request errors without touching stored images.

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.

Antigravity provider for DeepSeek Harness: resolves the OAuth client from the locally installed official desktop app and supports multiple accounts with quota-based automatic switching.

Adds a Speed setting (Standard, Fast, Ultrafast) for OpenAI and Codex models on a patched dsh build: the pinned dsh 0.2.1-alpha.2 web client needs the Model / Effort child-slot patch, the host pi-ai must be 0.87.1 or 1.1.0 (other versions refuse to load), and the terminal controls need a patched dsh-TUI. Then it appears in the Model / Effort menu; in dst /speed opens a picker or sets the speed directly, and the current speed shows in the bottom bar. The choice is kept per session, provider and model and is sent as the request service tier.