Plugins

Automatic Flash/Pro routing for DeepSeek Harness subagent calls: light tasks run on a cheap model (e.g. V4 Flash), heavy tasks stay on the strong model — with per-call overrides and /delegate session modes.

Runs reproducible DSH profile and patch experiments with guided case authoring, assertions, Web comparison reports, recovery, and policy gates.

Token usage and cost metering for dsh: per-session and aggregate statistics with configurable pricing and reports.

Acceptance-criteria-driven goal completion: locks verifiable criteria, tracks task dependencies, and gates completion behind formal reviewer validation (18 MCP tools).

Verification gate for DSH agents: every acceptance criterion must be backed by server-stamped real tool evidence before the completion gate lets a goal through (advisory audit, enforce gate, durable permits).

Delegates a self-contained coding task to the local Claude Code subscription through the official Claude Agent SDK, and adds a session tab with live output, cost and remaining quota.

DeepSeek Harness plugin: import models with capabilities from an OpenAI-compatible /v1/models listing (gateway capability extensions) and edit per-model capabilities (thinking, vision) on its settings page

A QR-code button above Settings that lets phones connect to the web UI through an auth-gated reverse proxy.

Transport-level authentication gate for the DeepSeek Harness Web GUI with server-side sessions, HttpOnly cookies, IP-based login throttling, and an scrypt password CLI.

Weekly digest generator: aggregates git commits, DSH session activity and daily memory entries into a Markdown weekly report, read-only.

Discord-native remote controller for DeepSeek Harness: drive persisted coding-agent sessions from DMs, mention-gated guild channels, and threads, with slash commands, progress updates, approvals, structured questions, and image and text attachments.

Unified messaging gateway for DeepSeek Harness: 27 IM platforms (Telegram, QQ, WeChat, Discord, WhatsApp, Feishu and more) in one plugin, with QR scan-to-authorize, per-platform workspaces, slash commands, and a web setup dialog.

Desktop (native) control for dsh as a stdio JSON-RPC bridge over a Python core: capture the screen, inject pointer/keyboard input, drive accessibility-tree actions, and gate high-risk operations behind required confirmations.

Gives an agent its own email inbox through AgentMail — eleven tools for sending, reading and labelling mail, plus inbound mail delivered into one harness session per email thread, with an approval gate and recipient allowlist on everything outbound.

Native per-workspace memory on the harness's own seams: facts and a bounded always-on profile on the storage-domain JSON unit, approval-gated writes with `(sessionId, seq)` citations, deterministic recall plus session-query FTS over past sessions — no external server, no custom SQLite.

Real-compute verification gates: math simplification, logic proofs, regex validation, E-prover FOL, state-machine checks, and code repair, backed by Go-compiled binaries with prebuilt Windows executables.

Bounded, privacy-safe local scanner for Codex, Claude Code, WorkBuddy, and CodeBuddy metadata: atomically aggregates keyed evidence and deterministically advises built-in DSH presets and optional capabilities—without content retention, network access, or preset mutation.

Bounded local memory with CBDC authority gating: SQLite + FTS5 claims, scoped recall with explainable use/verify/ignore decisions and a full audit trail, /memory commands, ≤3-claim/1200-char injection per call, no extra model call.

Run the Jev MCP server inside DeepSeek Harness. Typed judgment tools (verify, screen, find, review, gate, and the rest) register as mcp__jev__*. Provider keys stay in DSH credentials. Jev is not a chat model.

Pre-install security gate: quarantines each plugin, skill and subagent bundle with install scripts disabled, scans it with Cisco Skill Scanner, SkillScan, Cisco MCP Scanner, OSV-Scanner, Trivy, Gitleaks and optional SARIF tools or an LLM judge on a DSH model, then passes, asks for review or blocks it. Needs the install-gate hook proposed in deepseek-ai/deepseek-harness#12999.

Remake a viral short video end to end in one workbench: source clip, character sheet behind a human approval gate before any paid render, AI face swap, frame-sample QA and the final cut, plus a media library, a three-track timeline, and any RunningHub workflow connected by its ID.

Pre-execution review gate for file writes, edits and deletions: the tool call is held before it runs, a right-sidebar panel shows the computed diff (unified, or source versus result side by side), and nothing is written until Accept is clicked; pwsh commands that create, overwrite or delete files through a literal path are held the same way, and a command whose target is only known at runtime shows its raw command line instead of a diff.

Cross-computer Harness Agent messaging and task handoff; pairing, binding and the receiver's existing local authorization remain required. Not open source (UNLICENSED evaluation build, not for production); requires Node.js 24+, tested only on macOS arm64 and Windows 11 x64 build 22621+. Its last full install-and-run test was on DSH 0.2.1-rc.2; on 0.2.2-rc.1 only the version gate has been checked.

AlertManager panel in the DSH sidebar: reads multiple AlertManager HTTP API v2 instances in parallel, aggregates their alerts into severity-colored cards filterable by search, state, severity and instance, and dispatches any selection into a new session in a chosen workspace; the instance-grouped alerts become the first message to the agent, which then starts handling them immediately. Supports per-instance auth and TLS, auto-refresh, and a configurable dispatch prompt.