Agent collaboration · DeepSeek Harness plugins — Page 221

Complaint-driven competitor research: enumerate competitors, extract features/pricing/positioning plus cited user complaints from reviews, forums, and issue trackers, cluster the complaints, and deliver a comparison table with copy-worthy and skip lists.

computer use：让我自己看屏幕、自己操作鼠标键盘，从而独立完成「开发→验证→迭代」闭环

Spend the context window like a budget on long tasks: load files by graduated reads (head first, sections on demand, targeted search before full reads), store discovered facts to a notes file the moment they are established so nothing is re-derived, and re-state the running task plan at checkpoints — the difference between finishing a long task and drowning at 80%.

Local-first continuity plugin for Codex and DeepSeek Harness that preserves source-aware task invariants across lossy context transitions.

Lint your repo for chain-of-thought leakage — the session-transcript residue AI assistants leave in docs, comments, and commit-adjacent prose. Ships the cot-trim skill as an installable DeepSeek Harness plugin.

1-Line AI Agent Generator for DeepSeek Harness (dsh)

What will your agent's context cost before it runs? Audit instruction files, skills, MCP tool schemas and memory per source — with duplicates, always-on versus on-demand costs, and a budget you can enforce in CI.

Lightweight decision records for any project: capture decisions that are expensive to reverse or re-litigate — decision, alternatives, the why, reversibility — append-only, supersede instead of rewrite, kept where the project lives so the next session inherits the reasoning, not just the conclusion.

「替我审核」(auto-review) permission preset for DeepSeek Harness: a fourth permission mode beside read-only / workspace-write / danger-full-access. Escalation asks run through a three-tier funnel — deterministic whitelist (zero-token fast approval), deterministic blocklist (zero-token interception), and an independent reviewer model for the gray zone. Safe operations auto-approve, unsafe or uncertain ones fall through to the human approval prompt, and the reviewer model is user-switchable in General settings at runtime. Fail-closed by construction.

Live DeepSeek API cost/usage/balance dashboard for the DeepSeek Harness (DSH) Web UI: a second composer stats line plus per-session cost tracking — shows on/off-peak (Beijing-time windows), session cost (official CNY pricing), ~¥/min burn rate, account balance, and current model. 在 DSH 内实时展示 DeepSeek API 用量与费用：输入区常驻统计行，展示空闲/高峰时段（北京时间）、会话成本、每分钟烧钱速率、账户余额与当前模型。

Deepin/UOS desktop shell for the DeepSeek Harness (dsh) agent runtime — community-maintained Linux build of the official desktop shell.

An Electron-based desktop wrapper for deepseek-harness.

An unofficial desktop wrapper for DeepSeek Harness.

Community visual workflow and multi-model evaluation plugin for DeepSeek Harness

Hash-anchored UTF-8 file reading and atomic editing tools for DeepSeek Harness

DSH Lattice V1 aggregator: Cordis-style plugin registering ctx.lattice, orchestration engine, mailbox, agent_message shim, config loading, status projections, examples and docs.

Tiny native macOS shell for DeepSeek Harness — 132KB Swift+WKWebView, zero Electron

A terminal UI (TUI) for DeepSeek Harness — an opencode-inspired chat client that boots as a dsh profile app plugin.

DeepSeek-PA (DPA)：基于 DeepSeek Harness 的 Windows 桌面端。保留万物皆插件（Cordis）架构，Codex 风格界面，支持系统托盘常驻、开机自启、全局快捷键唤起。

deepseek-voice-mode browser surface: power toggle, live pill with captions and thread cards, push-to-talk

DeerFlow 2.0 (SuperAgent Harness) as a DeepSeek Harness (dsh) plugin — a Lead Agent that decomposes tasks into sub-tasks and dispatches them to pluggable skills (research, data-analysis, slide, web, image, video) with web search/crawl, sandboxed Python, reliability middleware, and cross-session memory, packaged per the deepseek-harness plugin spec.

Deliverable version discipline: files that left your hands are immutable, dates beat the word final, one authoritative delivery directory per project, and a ledger tying filename to date to count to recipient — so nobody ever asks which FINAL_v3 is the real one.

Pre-send verification for deliverables: opens the real file to verify row and dedup counts, scans for placeholder residue, diffs incremental files against their stated baseline, and logs the batch in a ledger with candidate vs sendable counts kept separate.

双 Agent 交付协作工作流（loop 模式）的 DeepSeek Harness 移植版：/delivery-review 命令 + 审查材料自动生成 + fixer/reviewer/planner 隔离子代理。基于 DSH 原生子代理消息通道（不使用 git notes）。