DSH Plugins Marketplace

DSH Plugins

Plugins

/

Sessions & Messages

/

dscode

d

dscode

Identified★ 1k

A DeepSeek coding agent harness: persistent shell, Ultra subagents, auto approval, Chrome MCP and session telemetry

DSCODE

❄ DSCODE

Write code in your terminal. Plug scripts into a live session. Hand tasks between agents.

Watch the 90-second DSCODE demo

macOS 14+ Linux glibc Bundled Node.js DSH License Release Stars Discussions

English · 简体中文

Core features · What's new · Quick start · Commands · Changelog · Docs

DSCODE is a terminal coding agent for macOS and Linux, built on DeepSeek Harness. Dedicated file tools read and edit code with file-version checks; a persistent shell handles searches, builds and tests. The TUI, the CLI and your scripts all share one session runtime instead of each starting their own. It installs as a pinned, reproducible harness—DSH dependencies, TUI and plugins are versioned and verified together.

Most coding agents work alone. DSCODE is built on the opposite assumption: sessions on your machine are visible to each other, so one can hand a task over, another can review the diff, and an independent reviewer decides approvals from your instruction rather than from a rule table.

See it in 60 seconds. Three things most coding agents cannot do, and where to look:

In one terminalWhat it shows
/btw why is the cache cold on the first turn?A side question runs in its own read-only child session and answers in a panel; the exchange never enters the main conversation.
dscode send <session-id> --steer "review the change in parser.ts and reply"Work handed to another session on this machine; it can read your transcript, answer, and hand the result back.
/permission auto-review · /review-usageApprovals decided by an independent reviewer from your instruction, with what it allowed and what it cost.

The demo guide summarizes DSCODE features, outlines a five-minute Desktop walkthrough, and retains the 90-second terminal shot list and recording commands.

🧭 Core features

FeatureWhat you get
Agentic coding loopNumbered file reads, exact edits and atomic writes with stale-version protection and diff metadata, plus a persistent shell for search, patches and tests. Project instructions, skills, plan, goal and hooks are wired in.
Session bridgeStart a task in the TUI, then add requirements, read output or subscribe to progress from another terminal (dscode sessions, send, read, watch). Every source enters the same runtime and context, and readers never take the session write lock.
Agent-to-agent tasksThe agent finds, reads and messages other sessions with list_sessions, read_session, send_session and reply_session, choosing queue, steer or defer. A persisted mailbox, retry de-duplication and a finite budget bound message loss, double processing and wake-up loops.
Session cardsEach session advertises its project, workspace and the topics of its last five user requests—enough to pick the right collaborator without reading its transcript. Cards describe what the user asked for, not conclusions.
Cross-session memoryReusable experience is extracted in the background and retrieved together with its workspace and source messages. Memory can be disabled per session or globally, and its model usage is tracked separately.
Effort and sub-agentsUltra uses the model's max reasoning and decides how far to investigate, delegate and verify; below Ultra the agent can still delegate, but only when a task clearly warrants it. Parents pick a separate effort per child, and children that edit can work in isolated Git worktrees created from a clean HEAD.
Independent reviewAfter a code change passes its relevant checks, the agent sends the Git diff—or, outside a repository, the changes since a snapshot taken when the task began—to a separate read-only model and fixes concrete findings before ending the turn. /review runs the same reviewer by hand, scoped to the staging area, a base branch, a commit or a path.
Non-interactive runsdscode exec "prompt", or git diff | dscode exec "review this", runs a full turn in scripts and CI: the reply streams to stdout, tool activity and the session id go to stderr, and the exit code reflects the turn. --json and --resume are supported.
Terminal UXSix interface languages, select-and-copy text, scrollback through history, visible tool failures and file-change summaries, expandable diffs, an input area pinned to the bottom, a sub-agent overview, and footer TPS / context / cost / cache figures. A bordered welcome card and full-height layout keep input at the bottom from startup. Default themes use aligned prose and restrained colours; Prismatic and Rainbow add decorative animation. The activity line shows the current target and tool time; verbose mode reveals thinking and calls. Cached balances retain currency and distinguish account funds from key limits.
GuardrailsA workspace-write sandbox by default and Auto permission review by an independent model, switchable to human approval. Ultra grants no extra permissions and Computer Use keeps human authorisation.
Agent emailA local [ToAgent] inbox shared by all sessions: Enter steers a message into the live session as user-selected context, and with Gmail configured the agent can send mail through the same approval flow.

🆕 What's new

0.7.46 — Self-contained packages for 64-bit glibc Linux (x64 and arm64). The same curl installer, dscode update and direct downloads now work on Linux; install bubblewrap for command sandboxing. See the release notes.

0.7.45 — The Antigravity provider recognises Gemini 4 Argon and selects its low/medium/high reasoning with the model's thinkingLevel instead of a token budget, which Argon rejects. Access is limited by Google to eligible accounts and live Argon generation is not yet verified. See the release notes.

0.7.44 — Adds native OpenAI/ChatGPT and community Antigravity Gemini providers, file-version-checked read/edit/write tools, a clearer full-height terminal layout and reliable non-interactive persistent Bash commands. OpenAI speed selection reports the server's actual tier; Antigravity account access still needs live verification. See the release notes.

0.7.42 — Fixes whole-package updates exceeding the archive-listing buffer. If you installed 0.7.41, rerun the curl installer below once; subsequent releases use dscode update normally. See the release notes.

0.7.41 — One self-contained terminal package includes its own runtime and native dependencies. Installation and updates no longer depend on npm or Hub; user data is separate, and an existing DSH installation stays independent. See the release notes.

0.7.40 — Clears cached balances when credentials change or are removed, preserves currency and account/key-limit labels, and adds fresh /status queries in the TUI and ACP. ACP also filters configured model choices and reports failed turns with retry guidance. See the release notes.

0.7.39 — Fixes DeepSeek provider startup in the published bundle. Upgrades the terminal to Harness 0.2.0-rc.2 and Computer Use 0.3.3, restores native reminder delivery and the /schedule catalog, and adds opt-in Atomic Mail MCP setup. Unconfigured or disabled Atomic Mail adds no tools or server instructions to model context. See the release notes.

0.7.36 — dscode acp connects ACP clients to the DSCODE agent preset, configured providers and client tool approvals. Install the npm latest launcher and configure your client with command dscode and arguments ["acp"]. See the ACP guide for inherited protocol limits and the release notes.

0.7.35 — Removes bundled Hub functionality, scopes Desktop browser tools to DSCODE sessions, and aligns settings and activation guidance with Harness. Existing Desktop models/accounts remain usable; TUI Browser Use is retained. See the release notes.

0.7.32 — OpenCode Go now signs in only with an OpenCode account: /opencode login shows a code, opens the console and signs in in the background. The login renews itself and sends requests to the inference endpoint the console names for your organisation. The console API key from 0.7.31 is no longer read, so run /opencode login once. The footer shows the subscription's usage of its five-hour, weekly and monthly caps, and when an exhausted one lifts. The consent page names the OpenCode CLI, whose login client DSCODE uses.

0.7.31 — OpenCode Go joins /provider: /provider opencode-go stores the key from the OpenCode console and serves Go's chat-completions models — DeepSeek, GLM, Kimi, MiMo, LongCat, Hy and Space Bunny — with the reasoning efforts each one accepted on the live gateway, the session id Go asks for, and reasoning passed back across tool calls. A Go session searches the web through Exa, as OpenCode does, with no extra key. Qwen, MiniMax and Go's Grok and GPT models use other protocols and are not offered yet.

0.7.30 — /delegate <task> turns the main agent into a coordinator: it plans the task on a board with priorities and dependencies, runs ready parts as child agents in isolated Git worktrees, verifies each result and merges the accepted ones in dependency order, staged and uncommitted. /delegate-dashboard shows the board as a colour-coded kanban — pending, running, verifying, complete. Parents now run up to five children at once below Ultra and twenty at Ultra, and the welcome snowflake is redrawn.

0.7.29 — /account signs the DeepSeek route in with a DeepSeek account through the system browser: the grant is stored on this machine, a signed-in machine needs no DEEPSEEK_API_KEY, and the command also reads the account identity, recharge and bonus balances, and signs out. The profile composes a loopback callback server for it, which carries a route only while a sign-in runs; DSCODE_ACCOUNT_LOGIN=0 removes it. Two composition fixes came with it: the host-plane workflow engine is disabled again after an upstream rename silently voided the patch that disabled it, and the redundant PTC runtime row is gone now that the base composition ships one.

0.7.28 — The DSH runtime moves from 0.1.5-rc.2 to 0.1.7-alpha.2, bringing sandbox escalation with justification on the retry shell, durable image offload, MCP resource tools, subagent permission inheritance and compaction retry recovery. The reviewed permission preset is renamed auto → auto-review, because DSH now reserves auto for its own integration: use /permission auto-review, dscode exec --permission auto-review and permission: auto-review. Computer Use keeps live activation and its native helper, but a resumed session regains its execution tools only once the computer-use skill is loaded again.

0.7.27 — Fixes fresh Hub installation failing after Harness composition generates its empty cordis.yml. The launcher carries a narrowly patched Hub CLI, and release verification now checks the native locked installer, launcher first start and Hub doctor.

0.7.26 — Triggers now support fresh or persistent sessions, durable cron and delay jobs, and supervised script sources. Manage them from /trigger, the CLI or agent tools with the same workspace and approval boundaries. The TUI stays in DSCODE mode and model switches reset unsupported effort values to the target default. TPS uses API-reported tokens with request-count exponential smoothing; runtime metrics and skills sit on the left, while cost, DeepSeek balance, peak/off-peak and cache align right or move together onto a third row. OpenRouter content-policy rejections now explain why the turn stopped.

0.7.25 — Running the installer again now keeps a tar installation current instead of refusing: an older one is updated through its own dscode update (state migrated, the old tree kept as .dscode-backup-<version>), an equal or newer one is reported and left alone, and anything the installer did not create is still refused with the path it found. A Homebrew formula joins the install paths — brew tap qiz029/tap && brew trust qiz029/tap && brew install dscode installs the same launcher, and there dscode update moves only the Hub profile because brew upgrade dscode owns the launcher.

0.7.24 — An event can start a session unattended: trigger definitions live in <state>/triggers/ and <workspace>/.dsh/triggers/, an event carries only data under a required eventId, and dscode trigger / /triggers run and inspect them, with trigger install writing a launchd LaunchAgent — a failed run notifies nobody yet. /goal[20] sets or re-caps the goal's round cap from the command plane. On OpenRouter, MiMo V2.6 (pro, flash and ultraspeed) joins the models that pass thinking back, and /model now offers a curated 28-model list — the five tuned labs plus the flagship line of Anthropic, OpenAI, Google and xAI — instead of all 373 tool-calling models, while a session already on a trimmed model keeps running.

Every release is listed in the changelog; the 0.7.29 notes have the long version.

🚀 Quick start

Requires macOS 14+ or 64-bit Linux with glibc (x64 or arm64; musl distributions such as Alpine are not supported). On Linux, install bubblewrap (apt install bubblewrap, dnf install bubblewrap) so commands run sandboxed; without it DSCODE falls back to Landlock and refuses sandboxed commands when neither is usable. Browser tasks also need Google Chrome, and repository/patch operations use Git. No system Node, npm, pnpm or DSH installation is required.

Install with one command:

curl -fsSL https://raw.githubusercontent.com/qiz029/dscode/main/install.sh | sh

The installer detects macOS or Linux and the processor architecture, downloads the latest self-contained release and verifies its SHA-256 checksum. It installs the complete application into ~/.local/share/dscode-app and links ~/.local/bin/dscode. Then run dscode in your project directory. If ~/.local/bin is not on PATH, add this line to your shell configuration and run it in the current terminal:

export PATH="$HOME/.local/bin:$PATH"

The package includes pinned Node, Harness, TUI, plugins and native dependencies. First launch never installs its runtime from npm or Hub. Your existing dsh command and ~/.dsh stay untouched; DSCODE keeps its data separately in ~/.local/share/dscode-state.

Update directly:

dscode update             # upgrade to the latest self-contained release
dscode update 0.7.46      # install an exact self-contained release

Exit running DSCODE sessions and schedulers first. The updater verifies and replaces the whole application, preserving settings, credentials and sessions; a failed replacement restores the old program. In the TUI, /update schedules the same updater after you exit. Re-running the installer also updates an installation it owns.

Coming from an older installation? Legacy npm/Hub versions need a one-time migration; their old dscode update does not switch to this distribution. The installer refuses to overwrite a dscode command owned by another installation. Homebrew-managed packages use brew upgrade dscode; the maintained tap installs the same self-contained release archives. See Distribution for migration steps. The commands above require a published self-contained release and do not fall back to legacy packages.

Prefer a direct download? Get one package from Releases: darwin-arm64 for Apple Silicon, darwin-x64 for Intel Macs, or linux-x64 / linux-arm64 for Linux. For example:

mkdir dscode-app
tar -xzf dscode-0.7.46-darwin-arm64-standalone.tar.gz -C dscode-app
./dscode-app/dscode

Keep the extracted directory together: dscode uses the private runtime and libraries beside it.

Enter /login and paste your DeepSeek API key into the hidden input: it is stored locally in ~/.local/share/dscode-state/credentials.yaml with 0600 permissions, shared across projects and installed versions, and never sent to the agent. A DEEPSEEK_API_KEY environment variable takes precedence. To use OpenRouter, enter /provider openrouter: DSCODE asks for your OpenRouter key (stored the same way, or taken from OPENROUTER_API_KEY) and switches the session to the DeepSeek model you were on; /model then lists a curated set instead of the whole catalogue—the current flagship models from DeepSeek, GLM, Kimi, Qwen and MiMo, plus the flagship line of Anthropic, OpenAI, Google and xAI—and typing searches it. Those five are tuned and tested; the Western flagships work on a best-effort basis. A model outside the list still runs when a session is already on it. /provider deepseek switches back, and /login openrouter replaces the key. An OpenCode Go subscription works the same way: /opencode login signs in with your OpenCode account, then /provider opencode-go offers Go's DeepSeek, GLM, Kimi, MiMo, LongCat, Hy and Space Bunny models. Use /model to pick a model or another provider, and /effort to adjust reasoning effort; the default route is deepseek-official/deepseek-flash. The TUI checks GitHub for a newer release at startup and /update upgrades the whole installation after you exit.

Use Antigravity with /antigravity login, then /provider antigravity, to call Gemini through your Google account while keeping DSCODE's tools and permissions. This community integration uses the model backend directly; account access and quotas are determined by Google.

Use OpenAI / ChatGPT with /openai login and /provider openai, or select API-key billing with /openai api-key. Choose reasoning with /effort and speed with /speed standard|fast|ultrafast. OpenAI conversations compact at 272,000 estimated input tokens, earlier for smaller models.

Use /provider custom to add a self-hosted or private API with Chat Completions, Responses or Anthropic Messages. Configure the endpoint and optional key, discover or enter models, set their context windows, and test streaming tool calls. For a model that supports vision, enable Input → Text and images to send screenshots and image attachments; the default is text only. Saved services appear by name in /provider; Enter switches to their model. See Custom model providers.

Six interface languages, English by default. /language opens the picker and /language ja switches straight to a locale; 中文, 日本語 and 한국어 are accepted as names too. The set is English, 简体中文, 繁體中文, 日本語, 한국어 and Español. The choice is stored per machine in $DSCODE_HOME/tui/language.json (under ~/.local/share/dscode-state by default), and DSCODE_LANGUAGE=es overrides it for one run.

dscode --continue                 # continue the last session
dscode --resume SESSION_ID        # resume a specific session (it runs in the folder it was created in)
dscode --cwd /another/project     # work in another directory
dscode trigger list               # event-driven runs defined in this project (run|fire|install|log)
dscode doctor                     # analyse recent logs and session traces
dscode --version                  # print the DSCODE version

The self-contained package uses its own pinned runtime and does not reuse or modify an existing DSH installation. Browser tools load on demand: ask the agent to use the browser, or run /browser start. CSS rule inspection uses snapshot element IDs and requires global and per-site Developer grants. Dedicated persistent profiles, existing-Chrome connections, experimental extension sharing of selected tabs, tab retention and the experimental Desktop sidebar preview with point annotations and site permission controls are described in Browser use. While visible and idle, the sidebar synchronizes permission changes made elsewhere, clears affected previews and keeps annotation drafts. The preview has been verified in the shared Web client and the official macOS Electron application using an isolated local fixture. Desktop tools load progressively through a skill; screenshot understanding needs an image-capable model. The MCP bridge supports tools and resources; Computer Use permissions are granted separately in macOS. Add other MCP servers in config/mcp.local.yml.

An experimental combined Desktop package adds the DSCODE preset, custom-model and account settings, browser preview, durable session communication, cross-session memory and opt-in scheduling to the official application. Settings → DSCODE → Schedules manages tasks and enables delivery while Desktop stays open. Build and local installation instructions are in Browser use. Native installation, upgrade, removal and reinstallation with scheduling enabled, plus rendered account and scheduling controls, were checked on macOS Desktop 0.2.0-rc.2; qualification limits are covered in Triggers. The package also includes an experimental native Computer Use adapter; window observation and input still await qualification with macOS Accessibility permission. Durable time context gives DSCODE agents arrival times, relay waits and previous turn endings while keeping ordinary chat clear. Namespaced Desktop diagnostics inspect the selected session, skills and Host MCP entries; matching user guides ship with the package. The Desktop email inbox previews shared local mail, connects IMAP or Gmail OAuth, and adds a selected message to a DSCODE session. Background sync is opt-in; outgoing mail keeps the existing approval and receipt rules. The Session usage sidebar shows DSCODE cost, context, cache, request speed and per-turn estimates. The Delegation board sidebar follows the selected main session’s tasks, dependencies, worktrees and verification evidence. The Desktop plugin is distributed separately from the terminal packages; terminal popups and terminal-only commands keep their existing UI.

The Desktop plugin uses the package name @toddzheng024/dscode-desktop. Version 0.7.35 removes bundled Plugin Hub UI, discovery tools and runtime dependencies; DSCODE remains listed in the Hub catalog. It targets Harness runtime 0.2.0-rc.2. In official Plugins → Add plugin, enter @toddzheng024/dscode-desktop@0.7.46, install, then choose Enable now. Fully quit and reopen Desktop after updating. Both preview and the default latest channel receive this verified preview release.

Settings → DSCODE groups setup, Models, Accounts and Schedules at the same level; Plugins → DSCODE · Community opens the same page. DSCODE browser tools are scoped to DSCODE sessions. Existing Desktop models/accounts work through the normal picker; optional provider settings use the native credential store. Enable Show coding view and Agent presets, then choose DSCODE for a new session after fully restarting.

Updates, data directories and source development

dscode update [exact-version] downloads and verifies the complete release, checks its bundled modules, then replaces the application as a unit. Exit running sessions first. A failure preserves or restores the old application; successful updates retain a sibling backup. There is no npm fallback. Packages installed with the new Homebrew formula use brew upgrade dscode to replace the same complete package.

Sessions, credentials and local configuration live outside the application at ~/.local/share/dscode-state, overridable with an absolute DSCODE_HOME. Startup isolates inherited DSH configuration and Node injection options; it does not replace the existing dsh command or ~/.dsh. Migration from an old installation is explicit; see Distribution.

Source development still uses Node and npm:

git clone https://github.com/qiz029/dscode.git
cd dscode
npm ci --ignore-scripts
npm run setup
npm start

⌨️ Commands

CommandWhat it does
/loginPaste the DeepSeek API key into a hidden input; saved under $DSCODE_HOME/credentials.yaml and loaded at startup
/accountSign in to a DeepSeek account in the browser instead of pasting a key; also its identity, balance and sign-out
/statusCurrent provider, model and credential balance; DeepSeek and OpenRouter supported, also available in ACP
/model, /effortChoose a model (type to search) or configure other credentials; models with four levels use a horizontal effort bar
/newStart a fresh session; the TUI is fixed to the dscode preset, including resumed conversations
/status, /doctorSession state and runtime diagnostics
/memoriesGlobal memory state, background usage, switches and cleanup
/sessionCurrent session id, its card, and the external access entry point
/permission auto-review, /permission askSwitch between automatic review and human approval
/review-usageExtra tokens and time spent on automatic review
/shell, /shell resetInspect or reset the persistent terminal
/browserBrowser sessions, profiles, extension pairing, site permissions, Developer mode, WebMCP, handoff and tab retention
/mcp, /skills, /hooksManage MCP, inspect skill sources, view or reload hooks
/plan, /goalPlans and long-running objectives
/agentsSub-agent tasks, status and current activity
/delegate <task>The main agent plans the task on a board with priorities and dependencies, dispatches ready tasks in priority order to child agents in isolated Git worktrees as child slots free up, answers their questions, verifies each result and finally stages what it merges
/delegate-dashboardKanban popup of the delegated tasks: pending, running, verifying, complete
/btwAsk a side question in a child session; the answer shows in its own panel and never enters the main conversation
/mailboxSession messages and deferred notes
/emailPress i to configure IMAP and an app password; browse [ToAgent] mail and press Enter to steer it into the session
/compactCompact the conversation context
/diff, /reviewInspect edits; /review asks an independent, read-only model to review a Git diff
/clearStart a new session with empty context; the old session stays resumable
/resume, /forkResume or fork a session

Shift+Enter inserts a newline, !command runs a local shell command, Ctrl+L clears only the screen, and Esc interrupts the current operation. The TUI's /help is the authoritative list, and commands and hooks covers the rest.

🔧 Permissions & configuration

The default is a workspace-write sandbox with Auto permission review, where applicable approval requests go to an independent model; /permission ask hands them back to you. Ultra adds no permissions. Scope, cost and limits of auto review →

Install methodSessions, credentials and statsLocal configuration
Self-contained package~/.local/share/dscode-state, overridable with DSCODE_HOME.env and config/ in the data directory
Legacy npm / Hub~/.local/share/dscode-hub, overridable with DSCODE_HOME.env and config/ in the data directory
Sourcerepository .runtime/repository .env and config/
Legacy tar.runtime/ inside the install directory.env and config/ in the install directory

config/hooks.local.json, config/mcp.local.yml and config/harness.local.yml are supported. Local configs, secrets and sessions never enter a release package. Skills are discovered in the target project's .agents/skills and .dsh/skills, plus an isolated user skill directory; /skills conflicts diagnoses same-name overrides. A project's .codex/hooks.json, .dsh/hooks.json and .claude/settings.json are layered on top of config/hooks.local.json by default; set DSCODE_PROJECT_HOOKS=0 to load the installation file alone. Skills come from the project's .dsh/skills and .agents/skills plus the user roots, and by default .dsh/skills, .agents/skills and .claude/skills are also read from every directory between the working directory and home, nearest first; DSCODE_SKILL_ANCESTORS=0 keeps discovery to the project and user roots alone. AGENTS.md/CLAUDE.md in those ancestor directories is folded into the workspace instructions.

🧩 Development

npm test                    # unit and contract tests (upstream dependencies isolated)
npm run check               # coverage, integration, TUI and packaging checks
npm run doctor              # deterministic local agent integration check
npm run dist                # build self-contained packages with private Node
node scripts/verify-standalone.mjs # qualify the package for this machine

make release runs version validation, the full checks, package builds, local installation verification and candidate archiving. CI also qualifies the x64 package on an Intel runner. make publish attaches terminal packages to the GitHub Release; Desktop retains independent build, verification and publication. node scripts/homebrew-formula.mjs generates a formula from the same archives and SHA-256 hashes; the formula still needs synchronization to the tap.

DSH is pinned to 0.2.0-rc.2, bundled Node to 24.21.0, the TUI is based on dsh-code@1.2.0, and Computer Use is 0.3.3. Runtime patches are applied during the build. See Distribution for the current flow and Legacy Hub distribution for retained npm/Hub tooling. Compaction evaluation lives under eval/.

📚 Documentation

DocumentContents
ChangelogEvery release, newest first
Non-interactive runsdscode exec, JSON output, resume and effort/model/permission flags
ACP clientsdscode acp, configured-provider model filtering, custom providers, client approval and protocol limits
TriggersEvent-driven runs with fresh or persistent sessions: cron, delay jobs, supervised script loops, /trigger TUI management, agent tools, the run log
Session bridgedscode sessions, send, read, watch against a live session
Session communicationAgent-side messaging, delivery modes, budgets and de-duplication
Session cardsProject, workspace and topic fields, and how they are derived
MemoryGlobal cross-session memory, background usage and switches
Persistent shell and UltraPresets, reasoning effort, per-child effort and worktree isolation
Computer UseNative macOS Desktop adapter, app grants, image evidence and qualification limits
Browser useBrowser setup, profiles, site permissions, Developer mode, WebMCP and task evaluation with DeepSeek or custom models
Plugin HubDSCODE/Hub separation and behavior of earlier published packages
Account loginSigning the DeepSeek route in with a browser account instead of an API key: the flow, the loopback callback, the limits
AntigravityGoogle account sign-in, Gemini models, conversation replay and connection limits
OpenAI / ChatGPTChatGPT login, API keys, Fast/Ultrafast speed and 272k compaction
OpenCode GoRunning a session on an OpenCode Go subscription: the account login, footer usage, the models and their reasoning efforts, what is not supported yet
Custom model providersConnect self-hosted and private APIs: protocols, model discovery, context windows, credentials and tool-call testing
Auto reviewScope, cost and limits of independent permission review
EmailIMAP setup, the inbox panel, send_email, aliases and opt-in Atomic Mail MCP (no context loaded until enabled)
Skills and workspace instructionsDiscovery scopes, the ancestor mode and instruction files
TUI commandsCommand reference and hooks
Session metricsDefinition of the footer TPS, context, cost and cache figures
Demo guideDesktop feature summary and walkthrough draft; 90-second terminal shot list and recording commands
npm + Hub distributionBundle, Hub release and launcher pipeline
Package distributionSelf-contained packages, updates, isolation and migration
VerificationWhat the maintained checks cover

Design records do not repeat the guides above:

DocumentContents
Session messaging designRationale, boundaries and fixed limits behind agent-to-agent tasks
Cloud web app hostDesign baseline and trust model for browser access; not implemented
TriggersWhy the trigger mechanism is shaped this way: the ingress, the supervised scheduler, the run log as a contract
VerificationWhat the maintained checks cover, and what cannot run inside a session
MaintainabilityWhat checks the forked terminal, and what stays unverified
Upgrading the vendored terminalTaking a new upstream dsh-code release: the merge procedure, and what only a smoke test catches
ReleasesLong-form notes for every release, newest first
Context handoffDevelopment checkpoint for a fresh session; not user documentation

🤝 Contributing

Issues and pull requests are welcome. Run npm run check before sending a change, and keep self-contained package verification separate from legacy npm/Hub tooling.

📄 License

MIT © 2026 Todd Zheng


Built on DeepSeek Harness, DSH-Code and DSH Plugin Hub. An independent community project, not an official DeepSeek product.

Comments

Loading…

From the same category

billion-context

by ranxianglei

基本稳定可用 100K tokens is enough. Universal context-compression proxy for ALL AI coding agents,10w上下文足矣

Sessions & MessagesManifest valid

★ 1.3k

↓ 619.8k/wk

MIT

TypeScript

Oct 15, 2026

dsh plugin --profile web add billion-context

by Han-1413141

DeepSeek Harness session cost meter plugin: session/daily cost, budget, history, OpenCode Go quota, official & custom-provider balance, Codex-like token heatmap, peak/off-peak pricing with pre-switch

Sessions & MessagesManifest valid

★ 420

↓ 27.8k/wk

MIT

JavaScript

Oct 14, 2026

dsh plugin --profile web add dsh-cost-meter

by Nwflower

Import 14+ external agent chat histories (Claude Code, Codex, ChatGPT, Cursor, Gemini, Reasonix, opencode, ZCode, Grok Build, OpenClaw, Pi, Hermes, Kimi CLI, DSH) into DeepSeek Harness as resumable se

Sessions & MessagesManifest valid

★ 231

MIT

JavaScript

Oct 13, 2026

dsh plugin --profile web add dsh-chat-import

by Totoro-qaq

DeepSeek Harness plugin for previewable cross-preset session migration. Fixed-schema handoffs preserve state, source-model intent, and unresolved images; the original session stays untouched.

Sessions & MessagesManifest valid

★ 163

MIT

JavaScript

Oct 14, 2026

dsh plugin --profile web add dsh-plugin-bridge

by MichengAI

DSH Archive Manager — 在 DeepSeek Harness 中安全查看、恢复和管理已归档会话 · Safely view, restore, and manage archived sessions in DSH

Sessions & MessagesManifest valid

★ 135

↓ 17k/wk

Apache-2.0

JavaScript

Oct 14, 2026

dsh plugin --profile web add @michengai/dsh-archive-manager

by Anionex

deepseek harness对话和代码状态回退插件 | DSH — rewind conversation and workspace state, powered by a persistent Change Ledger

Sessions & MessagesManifest valid

★ 133

BSD-3-Clause

JavaScript

Oct 1, 2026

dsh plugin --profile web add @anionex/dsh-turn-rewind