Plugins

Uses an isolated LLM review to approve or reject DSH sandbox permission requests.

Run the main session model on a local Claude Code subscription through the official Claude Agent SDK, taking over the llm/stream route for the claude-code provider, with token-level streaming, session resume, and DSH tools bridged in over MCP so they keep DSH sandbox and approval.

Antigravity-style run_command for DSH: short commands return inline within a configurable wait window (default 10s) while longer ones auto-promote into ctx.jobs for later collection and completion notices — executed via ctx.shell under the session sandbox and approval pipeline.

LLM approval answerer for sandbox escalations beyond workspace-write: a deterministic regex filter first, then a clean-context reviewer model; humans are asked only when it is unsure. Requires the bundled core patches.

Multi-folder workspace for DeepSeek Harness: pick several folders and create one workspace spanning them (directory-flow occupant + extended workspace/sandbox contract)

Envoltorio para el CLI de OpenAI Codex: ejecución en un solo disparo, revisión de repositorio y reanudación de sesión con un sandbox de solo lectura por defecto.

Barreras de autoprotección para DeepSeek Harness: interceptación de rutas protegidas, copia de seguridad antes de destruir, comandos de autorrecuperación de CLI y configuración predeterminada de sandbox estricta.

Modo de permiso de revisión automática al estilo Codex: añade un preajuste de revisión automática que aprueba automáticamente las escaladas de sandbox seguras, pregunta por las riesgosas o ambiguas y rechaza las operaciones críticas no confirmadas.

Sandbox multi-espacio de trabajo: concede acceso de escritura a todos los espacios de trabajo registrados automáticamente — añade uno y escribe en él al instante, sin configuración ni escalado de permisos.

Soporte de sandbox multiplataforma de Microsoft.

Escalation-only LLM reviewer: reviews sandbox escapes only, keeps the sandbox, fails closed.

Replaces the web profile's pwsh shell executor with a subclass that rewrites every command through `rtk rewrite` before execution for token savings, running the original command unchanged on any rewrite failure (no equivalent, missing rtk, timeout) and keeping sandbox semantics intact.

Approval-free writes into whitelisted directories — wraps the filesystem fence for write/edit tools and auto-answers bash sandbox escalations that touch whitelisted paths, with a settings GUI.

Maintained fork of dsh-approval-gate. A neutral operation whose confirmation count has reached the threshold auto-approves when the judge is unavailable, instead of prompting a human again; approver-facing explanations are generated in Chinese from the real sandbox mode, command and paths. Also carries deterministic hard-deny for credential exfiltration and system-path destruction, judge-input redaction, a judge model candidate chain, fingerprint-scoped allow rules, and an approval view with unified diff and one-click revert.

Use your DSH models in [Libre WebUI](https://github.com/libre-webui/libre-webui) for chat, sandboxed Work and usage tracking, with provider credentials kept in DSH. Requires Libre WebUI 0.37.0 or later.

让 dsh 在 Windows 上使用用户自安装的 Git Bash，并绕过其沙箱限制 | Enable dsh to use a user-installed Git Bash on Windows by bypassing its sandbox restrictions.

Approval-free writes into whitelisted directories — wraps the filesystem fence for write/edit tools and auto-answers bash sandbox escalations that touch whitelisted paths, with a settings GUI.

Keeps a redundant sandbox-escalation argument from failing a tool call. Escalating tools (pwsh, bash, write, edit) advertise the full sandbox_permissions enum, but DSH only accepts a level strictly wider than the one already in effect — a check its own source calls 'deliberately not a schema constraint'. A model that reflexively passes the argument therefore picks the level it is already at and the call dies before running with 'sandbox escalation to "workspace-write" is not strictly wider than this call's current "workspace-write" mode', costing some models a whole turn of retries. This plugin registers one tools/execute waterfall listener and, only on that documented rejection and only when the arguments really carried an escalation field, re-issues the identical call once without it. Safe by DSH's own documentation: the rejection precedes any execution ('nothing has run'), and the corrected call cannot match again, so the retry is loop-free. Reproduced and verified end to end — before: isError true, the command never ran; after: the command's real output, isError false, one tool/call and one tool/result. Zero dependencies.

Make a Git for Windows installation resolvable as `bash` inside the dsh host process, so the official bash shell stack (dsh-bash-sandbox + dsh-tool-bash) can run on Windows

Engineering work-tools for DeepSeek Harness: registers two model-facing Host Tools — a scoped `git` runner, and a package-manager runner (`dev`) that detects npm/pnpm/yarn/bun from the workspace lockfile and picks the right verb per manager. Both run through dsh's sandboxed shell and return structured stdout/stderr/exitCode, so the agent gets focused, reviewable results instead of raw bash, and adding another tool is a copy-paste of one register block.

Require a real, immediately preceding sandbox denial before DSH Bash can request wider permissions.

Patch-style file editing: one apply_patch tool accepting git/unified diff (default) and Codex apply_patch syntax (opt-in), applying every change all-or-nothing after a full read-only verification pass, with sandbox-aware delete and move and a dry-run mode.

RTK (Reusable Token Kompressor) tool for DeepSeek Harness — token-optimized CLI proxy exposing read/ls/git/grep/err/test/json and more through the standard shell sandbox.

Teaching board view: agent-generated self-contained HTML is projected into a sandboxed iframe as a board, annotated by hand (brush, four colours, eraser, undo), extended in place across several tool calls, and screenshotted back into the conversation as a real user message.