Plugins

Memória entre sessões para DeepSeek Harness via o domínio de armazenamento oficial: memórias persistem através de ctx.storage (tabela storageDomain) e são injetadas no system prompt através de uma seção registrada (memory:recall), com ferramentas memory_save / memory_forget para que o agente leia e escreva sua própria memória (maxMemories configurável).

Painel de gestão de atenção de tokens para o DeepSeek Harness: acompanhamento de uso e custo de tokens por tarefa/dia/semana/mês (acerto/erro de entrada, saída, raciocínio), preços de pico/fora de pico do DeepSeek, reconhecimento do tipo de tarefa e conselhos sobre quando trocar de sessão, escrever uma passagem ou comprimir o contexto.

Gerenciador de nível de esforço de raciocínio para modelos de API customizados, mais memória de preferências entre sessões: toggles inline low/medium/high/max dentro de Configurações → Modelos, com escritas atômicas de configurações.

Forma de pacote estático do portão de segurança de instalação de plugins do DSH: revisa cordis_define/cordis_run com uma política à prova de falhas, além de ferramentas de revisão/auditoria e um popup de aprovação no navegador (agree / agree+whitelist / reject; agree+whitelist grava a família de plugins em tru).

Controle deslizante de esforço de raciocínio em pixel art para DSH Web com arrasto contínuo de 0–100, gravações com limitação a 16 ms, confirmações com encaixe em nível, e um fogo de brasa WebGL que se acende conforme o controle se move.

Registra a skill remove-pdf-background-gray para DSH: clareia fundos de digitalização cinza/off-white em PDFs baseados em imagens, preservando a resolução, a geometria da página e as bordas de texto com antialiasing (regravação Flate sem perdas), por meio de um único script Python (pypdf + Pillow + numpy).

Conjunto de competências do agente de notas virais Xiaohongshu/RED: pesquisa de notas em alta, geração/reescrita de notas, verificação, análise de contas autorizadas, início de sessão por QR e publicação controlada.

Analisa código local e GitHub/npm antes do desenvolvimento para encontrar implementações reutilizáveis, avalia reutilização versus reescrita e verifica o acoplamento da arquitetura.

Memória nativa por workspace sobre as próprias costuras do harness: fatos e um perfil sempre ativo, porém limitado, na unidade JSON do domínio de armazenamento, gravações com aprovação e citações `(sessionId, seq)`, recuperação determinística e FTS de session-query sobre sessões passadas — sem servidor externo, sem SQLite personalizado.

Revisa todos os arquivos que o agente escreveu em uma sessão, a partir do cabeçalho da sessão, com hunks por arquivo e escritas recusadas contadas separadamente das alterações.

Bloco de escrita Markdown com escopo de sessão encaixado ao lado da conversa, com modos de edição/visualização, reescritas de IA baseadas em seleção, revisão de diferenças, desfazer e recuperação de rascunhos.

Kit de operação com foco em evidências: seis ferramentas somente leitura (catálogo de capacidades, planos de workflow em etapas, leitor de skills empacotadas, busca limitada em memória local, auditoria de releases do repositório, checklist de release) e cinco skills empacotadas — planeja e audita, nunca escreve remotamente.

Claude Code's auto memory in DeepSeek Harness: the index in every request, and tools to write and edit it

Second-model gate on the assistant turn, installed in two parts: copy engine/ into ~/.dsh/plugins/ and mount it by absolute path in the profile patch (the bundle itself only installs the settings panel), then switch off the default dry-run mode that only logs verdicts. Once active, a separate judge call scores the draft against the numbered criteria (0-10 each); any criterion under 6 fails the turn and sends it back for a rewrite, up to 3 rounds by default. The web panel edits criteria and thresholds, and every verdict is appended to an audit log.

The cyber-elixir of youth for the desktop client: fixes the `desktop welcome: Web authentication failed` (HTTP 431) launch failure — an Electron random-port launch lets `dsh-auth-*` login cookies pile up per port (70+ of them, ~229 B each), the startup request headers blow past Node's default 16 KB cap, and the client can no longer launch. Two fixes in one: a scriptable cleaner that removes leftover login cookies from the cookie database (Chromium SQLite) — explicit, offline, keeps a `.bak` copy, `--expired-only` narrows the scope — and a `patch` feature on the settings panel that fixes the launch port through the profile `cordis.patch.yml`, so the client always starts from a single port and cookies stop accumulating across different ones (rewrite `port` as a `!!js` expression to keep the `--port` flag in charge; a per-profile snapshot makes one click revert it). Read the README for usage. Interface, CLI output and market metadata in 9 languages. The purge runs from the bundled CLI (`dsh-cookie-cleaner --clean`) only while the client is fully closed; the in-app half is read-only. npm: `dsh-cookie-cleaner`.

A consent-gated personal profile store: people are indexed by name, entries by person and key, and no profile record is read before the current session is authorised. Sensitive entries come back masked unless the user authorises that tier separately; credential-like values are refused at write time. Ages and next birthdays are computed at read time, so asOf restores the answer for a past date, and facts that rot (height, weight, address) carry freshness warnings. Five profile_* tools plus a settings page to browse, edit, delete and revoke grants.

A memory plugin that forgets on purpose: short-term memories expire after 48h, long-term ones are rescheduled by Ebbinghaus/SM-2 curves and rarely-used ones fall asleep instead of bloating your context. Long-term writes stop in an approval center, so the agent cannot rewrite your persona on its own.

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.

Adds two enhancements to DSH ask_user_question choice cards: right-clicking an option expands a per-option note that is merged into the submitted custom answer, and a footer field asks a side question answered by a fork subagent from the session context, read-only, with no write-back. Keeps left-click, skip, paging and the countdown working as before, and only claims generic choice cards, not plan review or approval cards. Supports DSH 0.2.x from 0.2.0-rc.1 (verified on 0.2.2-rc.1) and Node.js 20+. Derived from buaaflyaway/dsh-question-note (MIT) and inspired by MichengAI/dsh-btw (Apache-2.0); do not install it together with dsh-question-note — both take over the same ask_user_question card.

Optimize prompts in the DeepSeek Harness composer: identify missing requirements, insert a sentence from 90 method cards, or compare and edit a full rewrite before you send.

A sticky-note wall in the DSH sidebar: nine paper colors, categories, search, archive, to-do checklists with stable item ids, and agent read/write tools.

Brings the rtk token-saving setup from Claude Code to DeepSeek Harness: commands rtk can rewrite are rewritten, and builds, tests and logs it cannot reach are compressed; git, cargo, npm and pytest are covered. Compressed output tells the model what was elided and how to recover it; when the sandbox cannot reach the rtk store the original is saved and its path given. Savings show without a command: a pill below the composer, with `/rtk stats` for detail. When a rewrite is unreliable the original command runs as usual: if rtk cannot run, or the returned command would drop arguments, nothing breaks. The rtk binary ships with the package, so mainstream platforms need no separate rtk install, unlike most other rtk plugins.

Qoder-style workspace mode and branch chips on the session row; picking a branch arms it, and the first message creates a dedicated git worktree and moves the session into it. Windows only: the write grant runs PowerShell Set-Acl; elsewhere Worktree mode reports an error and stays in the main checkout.

Multi-round deep research in the DSH sidebar: each round fans sub-questions out to parallel sub-agents, then continues from where the last round stopped and writes a cited report.