Agent collaboration · DeepSeek Harness plugins — Page 215

Cross-session relay for DeepSeek Harness: peer discovery and text delivery between independent dsh processes.

Web agent-heartbeat sonification for DeepSeek Harness: a calm lub-dub hum while agents work, ticks on streaming output, chime when a session awaits your input

Remote SSH terminal for the DeepSeek Harness web UI: floating terminal panel, dangerous-command guard (open/guarded/locked), confirmation workflow, and model tools (netshell_servers / netshell_run). Passwords live in the DSH encrypted credential store and never pass through AI sessions.

Evidence-based delivery for DeepSeek Harness: deliverable contracts, artifact-hash checks, and stale invalidation folded from the session log

Local-filesystem provider for dsh-evidence: built-in artifact validators (exists, nonempty, json, json-keys, html-basics) with sha256 hashing

Completion guard for DeepSeek Harness: turns agent/turn-stopping into an evidence check so "done" means "verified artifacts", not a model claim

Intent routing and run contracts for DeepSeek Harness: normalize fuzzy requests into a single-active-step contract folded from the session log

Automatic review decider for dsh-review-separation: risk-tiered whitelist rules that short-circuit reviewer subagents for safe work

Review separation for DeepSeek Harness: maker/checker orchestration over the subagent seam — independent reviewer verdicts folded from the session log, rejects stale the work

Model-facing evidence tools for DeepSeek Harness: declare_evidence, check_evidence, evidence_status

Model-facing intent tools for DeepSeek Harness: declare_intent, complete_step, intent_status

Model-facing review tools for DeepSeek Harness: request_review and review_status over ctx.reviewPolicy

Forgetting engine for AI agents — memory TTL, decay, eviction, audit. The opposite of memory programming.

Oh My Pi bash runtime ported as a deepseek-harness tool plugin: persistent shell session, output minimizer, managed background jobs, command interception

AI 生图 (image generation) plugin for the dsh web GUI: text-to-image and image-to-image through a configurable OpenAI-compatible endpoint, with multi-provider support, model discovery, and agent tool registration. 与 @xiaoyuink/dsh-image-vision 同系列的图像插件。

Cordis (DeepSeek Harness) plugin: normalized GitHub API tools for the agent — gh CLI by default, native HTTP fallback when gh is unavailable

mycel working modes: a switch_mode tool returning one selected mindset as an ordinary tool result, plus a shared core-behavior system-prompt section

mycel session state as a plain-text notebook: add_state / read_state / compact_state tools over mycel/state log events; survives compaction and resume, never auto-injected

把九种 IM 机器人和公网 AI Office 接入本机 DeepSeek Harness。 Connect nine IM channels and a public AI Office to a local DeepSeek Harness.

Cursor Agent for DeepSeek Harness: browser login, token refresh, model discovery, and Cursor's server-side agent loop

A Git worktree plugin for DeepSeek Harness that creates task branches, registers worktrees as DSH Workspaces, and opens isolated sessions.

DSH plugin: connect to Futu (富途) MCP via OAuth 2.1 and expose its tools to the agent; tokens persisted outside the workspace

DSH 插件:本地分流代理 + Web 设置页 —— 只有规则命中的被墙域名走上游代理,其余(国内/内网/未知)直连;上游地址、清单、超时等参数直接在设置里改,热生效、免重启,并带运行状态面板 | DSH plugin: local routing proxy with a Web settings page — only rule-matched (blocked) hosts use the upstream proxy, everything else goes direct; the upstream address, blocklists, timeouts and more are edited in Settings and applied live, with a built-in status panel

Claude Code subagent provider with configurable CLI executable path, forked from @deepseek-ai/dsh-subagent-claude-code