Plugins

Plugin do DSH Web para gerenciar sessões: arquivar/desarquivar, exclusão física garantida (arquivos transcript e varredura de ~/storages/*.json), reatribuição de workspace por arrastar e soltar e edição de mensagens por branch (edit/retry/reroll com linha do tempo de versões reversível).

Camada de validação de experiência para DSH: coleta sem tokens de sinais de falha de ferramenta/nova tentativa/busca, refinamento por IA em livros de habilidades verificados (verificação em três camadas), consultados no início da tarefa. Inclui 10 habilidades de teste e dados de benchmark (sucesso em tarefas complexas 100% vs 60% sem o plugin).

Roteamento multimodal transparente para modelos somente de texto: cada imagem em cada chamada de modelo é totalmente transcrita (OCR literal, dados, zonas de incerteza, reforçado contra injeção) pelo seu próprio compreensor multimodal, com nova observação focada via vision_relook e nova tentativa automática em falhas relacionadas a imagens. Sem endpoints incluídos, sem logins emprestados.

Repete automaticamente solicitações de API com falha para manter as conversas do DSH em andamento. Suporta códigos de status e códigos de erro configuráveis.

Pacote de proteção contra mojibake para o DSH: detecção de texto corrompido baseada em características antes da escrita com revisão e liberação (uma nova tentativa idêntica passa), mais auditoria de codificação do pwsh antes da execução (ANSI/GBK/UTF-16/Default). Windows e chinês primeiro.

Keeps a turn from closing on an unfinished reply: a stream guard turns silent truncation into an in-step retry, and a completion protocol asks the model to declare how each turn ended.

Read-only /btw side questions for DeepSeek Harness: forks the current session, injects the running turn's task, recent tool calls and partial answer so a mid-turn question has context, denies every tool call at execution time, and keeps a per-session topic history with follow-up, stop, retry and delete.

Sends desktop notifications when a DSH turn finishes or fails, and sends 继续 automatically after the model-retry budget is exhausted.

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.

Prepara arquivos antes de read, read_image e as ferramentas vision os abrirem. Reduz imagens acima de 2048 px ou 8 MB (uma nova tentativa em 1280 px no mesmo turno), converte BMP/TIFF para PNG e identifica uploads com rótulo incorreto examinando os bytes. Converte docx/xlsx/pptx, PDF com camada de texto, csv/tsv e html em Markdown sem programas externos.

Quando o limite de contexto é excedido e a compactação integrada desiste, compacta mais uma vez automaticamente e tenta de novo. Se falhar novamente, organiza a sessão original em um pacote handoff para continuar em uma nova. Sem trocar o mecanismo ou adicionar UI; a retomada da sessão exige acionamento explícito por padrão, e a original fica somente leitura.

Proteção contra falhas no processo: assume os handlers de erro fatal que encerrariam o DSH, identifica o plugin responsável pela stack e desativa somente sua linha na camada home patch, sem alterar profile ou a configuração provider dentro dele. A página de falha oferece tentar novamente, manter desativado ou desinstalar; sem causa identificável, abre uma sessão temporária safe-mode (somente lançamentos CLI e Web).

Recusa esperas em primeiro plano nos agentes DSH: comandos de espera (sleep, wait, tail -f, watch, ping sem -c, loops infinitos), ferramentas bloqueantes (wait_agent, job_output com wait: true), repetição do mesmo comando de verificação dentro de uma janela e comandos que exigem intervenção humana (sudo sem -n, ssh sem BatchMode, login, editores interativos, docker -it). Sugere uma alternativa acionável: executar em segundo plano ou encerrar o turno. Orçamento e janela de polling configuráveis; idioma da mensagem segue system (zh/en). Sem dependências de runtime.

Tenta novamente automaticamente solicitações de modelo com falha: intercepta cada erro e repete a mesma etapa após espera de 2/5/10/20 s, evitando que falhas do gateway e limites de taxa encerrem o turno.

Anel de cota do GLM Coding Plan no dock do composer: janelas de crédito de 5 horas e semanal, nível, atualização automática e nova tentativa com um clique. Requer assinatura e chave do GLM Coding Plan, além dos IDs de organização e projeto BigModel (a API de cota de equipe fica vazia sem eles).

Adiciona a opção nativa “Restart DeepSeek Harness” ao menu da bandeja do sistema do Windows e ao menu da conta do app DSH para desktop. Na inicialização, o plugin prepara um resources/app.asar com patch; um auxiliar separado faz backup do original e o substitui após o encerramento normal. O patch é reaplicado após cada atualização do desktop. Qualquer uma das opções reinicia o app normalmente na próxima inicialização. Apenas Windows desktop; outros hosts retornam 409 not_desktop.

Botão ao lado do composer que reescreve o rascunho para um prompt mais executável. Quatro presets, de polimento leve a expansão agressiva, e presets próprios. Troque preset/modelo de prompt/esforço de raciocínio no botão sem sair do composer. Só aparece com texto; mostra timer durante o ajuste e oferece nova tentativa se falhar.

Veja no DSH a cota real da assinatura ChatGPT/Codex. Lê `/wham/usage`: janelas de 5 h e semanais, contagem até redefinição, saldo de uso extra, grupos de limites adicionais e limite de revisão de código. Lista/resgata créditos de reset emitidos pelo servidor com idempotency key estável e segura para novas tentativas; aceita várias contas. Lê `/codex/models` para o catálogo real e fica vazio quando indisponível, sem lista embutida. Login de um clique em Settings (autorização no navegador ou device code, com alternativa de colar callback); credenciais no credential service do DSH. As chamadas reutilizam a rota Codex integrada `llm-pi-ai`, sem registrar nem ocupar uma rota provider. Se outro plugin possuir essa rota, resta apenas o console de cotas. Chip de cota restante junto ao composer.

Plugin de reparo de ACL do sandbox Windows para DeepSeek Harness: corrige a falha grantWrite (SetNamedSecurityInfoW Win32 5) e tenta o comando novamente sem exibi-lo.

Encaminha registros de conversas do WeChat para sessões DSH: um destino de compartilhamento do Windows captura o ZIP encaminhado; uma sobreposição GUI permite escolher workspace e sessão. O ZIP chega como mensagem do usuário com anexo, que o agente extrai e lê. Normaliza extensões pelos magic bytes (o WeChat pode enviar dados PNG/HEIC com nome .jpg) e preserva o lote se a entrega falhar para permitir nova tentativa. Requer Windows 10/11, Developer Mode ativado e instalador do repositório para registrar o destino.

Bridge IM para DSH: 15 adapters/12 plataformas. Adapta hosts 0.1.2-rc.1–0.2.1-rc.2 via capability probing (enum 12 tags, sem ramo por versão); ffmpeg só para voz Feishu. Aprovações/perguntas no chat de origem: deny-by-default, só iniciador, expira em 5 min. Imagem/arquivo conforme metadata do modelo host; devolve arquivos, retry limitado (envio incerto não é repetido). Tabs Tasks/Channels cedem o slot se outro plugin o possuir. Preset im-concise; remove GenUI fences/raw HTML de chats texto, svg→PNG. Allowlist de modelos. /relay envia resposta final por account ID/alias, sem adivinhar ambiguidades. Secrets só vão ao vault, nunca retornam nas respostas. npm: dsh-im-union.

Conecta o DeepSeek Harness ao serviço Hindsight seguindo as convenções do omp: banco compartilhado, tags de projeto automáticas, recuperação por projeto e global, e ferramentas de modelos mentais, duplicatas e repetição de operações em segundo plano com falha. Requer servidor ativo (padrão http://127.0.0.1:8888).

Observa falhas consecutivas idênticas de ferramentas e pode pausar antes da próxima etapa do modelo. Preserva os resultados originais e as repetições bem-sucedidas e mostra um relatório dentro do chat no DSH Web.

Recupera turnos DSH interrompidos: amplia o orçamento próprio de tentativas e mostra a espera como contagem regressiva ao vivo na mesma etapa. Mantém uma saída truncada no mesmo turno e pode abrir uma rodada opcional de continuação depois do fim; com o compositor vazio, o botão enviar inicia essa continuação.