Plugins

Fail-closed network recovery, recommended answers, and safe unattended continuation for DeepSeek Harness.

DeepSeek Harness (dsh) approval answerer backed by the agent-governance sidecar — every approval/request is decided by the deny-by-default gate, L0–L4 rules and the organizational audit chain.

Human-in-the-loop tools (`human_task` / `human_task_ready_check`): pause the agent until the user performs a real-world action or returns an observation (GUI operation, game testing, visual verification, hardware checks), gated by session consent and an AFK presence check, returning a structured JSON result via a web task dialog.

Manage iterative text annotation through DSH tools with rule versions, stale-result isolation, human review preservation and traceable exports; requires a separately installed Python backend.

Source-backed logic reviews with diagrams, code diffs, and resumable human feedback. The first activation downloads the official LogicMargin release (SHA-256 verified); on macOS and Linux this needs curl, unzip and shasum or sha256sum.

Human-in-the-loop self-improvement loop: one objective becomes an auditable work chain that iterates in child sessions while the main agent stays idle.

Collects the open questions of each AI turn into a right-side panel; you answer them, add notes, and send the decisions back to the conversation in one message.

Publish agent output to Reviso for human review, browse published documents from the sidebar, and find them again from the session that produced them. OAuth is host-mediated: Reviso tokens stay in DSH credentials and the browser only receives short-lived embed sessions. Requires a Reviso account or a compatible self-hosted Reviso server.

Host-side Feishu board for DeepSeek Harness: pushes a session's ask_user_question to Feishu DM as plain text and sends a quoted reply back to the exact session that asked, archiving unmatched replies instead of guessing. Its lark-cli wrapper also suppresses an identical notification resent within three minutes.

Lets a DSH agent ask the user a structured question without blocking the turn: the ask_interactive_question tool returns a pending batch id at once, the answer arrives later as an ordinary user message, and pending batches sit in a browser inbox panel with optional system notifications.

Computer Use enhancement plugin for DeepSeek Harness (DSH). 纯 JS 视觉差分、网格位置编码、长程任务闭环与人机共驾授权。

A human-in-the-loop curation workbench inside a DSH session: mark whole turns or single messages to keep verbatim or to summarize, rewrite user, assistant, tool-result and injected-context messages in place without a model call, then compress the marked ranges with model-written summaries or the official /compact. A local JSON backup is written before every write to the session.

Writes a model-generated annotation into every DSH escalation approval prompt — what the command does, what the model intends, how risky it is and why, and what each choice means — and keeps a searchable history of past approvals at /dsh-approval/history, with a diagnostics page at /dsh-approval/diag.

A browser that lives inside the DSH host: every session gets its own tab, the AI drives it with lazily published browser_embedded_* tools, and a human watches or takes over that same tab in an on-demand right-Sidebar tab — no GUI, no external browser, log

A browser that lives in the DSH host: every session gets its own tab, the AI drives it, and you watch or take over that same tab inside the DSH Web UI — logins, 2FA, QR codes and CAPTCHAs included.

Unofficial host-side ServerChan notifications for DeepSeek Harness: sends mobile alerts through WeChat (ServerChan Turbo) or the ServerChan³ app when approvals, plan reviews, or questions go unanswered beyond a configurable threshold, even with the browser closed.

Human-in-the-loop tools (`human_task` / `human_task_ready_check`): pause the agent until the user performs a real-world action or returns an observation (GUI operation, game testing, visual verification, hardware checks), gated by session consent and an AFK presence check, returning a structured JSON result via a web task dialog.