ParameciumU

Create a new monoU agent and connect it to Gateway. Use when the user asks to create/new/spawn an agent (本地或远程). Supports one-command local creation, remote deploy, Windows persistent startup, and skill/persona customization.

Minimal terminal coding harness. Use when the user wants coding-agent behavior: read files, run bash commands, edit files with exact replace, write new files; prefer read-before-edit, grep/find/ls for discovery; concise responses with clear paths.

定时任务管理：列出、创建、更新、删除、立即运行与查看状态。调度类型支持 at（一次性时间）、every（固定间隔）、cron（cron 表达式）。不依赖 openclaw；任务持久化在 .first_paramecium/cron/jobs.json。

工作区长期记忆的检索、读取与写入，持久化在 ./.first_paramecium 下。工具：memory_search、memory_get、memory_store、memory_recall、memory_forget、memory_sync。短期/多轮由 state.messages 提供；跨会话用 memory_store 或 write 写入后用 memory_search 回忆。

Minimal terminal and file harness. Use when the agent needs to read files, run bash commands, edit or write files; prefer read-before-edit; concise responses with clear paths.

Export solution content to a standard Word document (.docx). Use when the user says '输出word文档' / '导出Word' / '生成Word' / '下载Word'. Call generate_word_document with the current solution content, client name, and optional date.

Requirement interpretation and solution document generation. Use when the user uploads or pastes customer requirements (Word/Excel or plain text), or asks to generate/revise a solution document. Always follow the 6-chapter template; support multi-turn iteration from conversation history.

Minimal terminal coding harness. Use when the user wants coding-agent behavior: read files, run bash commands, edit files with exact replace, write new files; prefer read-before-edit, grep/find/ls for discovery; concise responses with clear paths.

Fetch URL content or search the web. Use web_fetch for a known URL; use web_search when you need to find current information.

Send a message to a connected channel (e.g. Feishu). Only available when agent runs with Gateway.

代码库发现与导航。提供 grep、glob、list、code_search（自然语言/关键词搜索）、apply_patch（V4A 多文件 diff），与 base_skill 的 read/edit/write 配合使用。

Manage scheduled tasks: list, create, update, delete, and immediately trigger jobs. Supports at (one-shot), every (fixed interval), and cron expression schedules. Two execution modes — agentTurn (LLM agent runs a turn) and systemEvent (no LLM, handled directly by the runtime). Use when the user wants to schedule something at a time or interval, view existing jobs, or manage cron tasks.

Complete toolset for interacting with the monoU Gateway. Discover topology (online agents, L3 nodes and their capabilities), query per-agent skills and cron jobs, delegate tasks to other agents, invoke L3 node capabilities (browser-node / sandbox-node via node.invoke, analogous to MCP servers), manage sessions, and push messages to connectors (e.g. Feishu). Use when the user asks about: who is online, what agents can do, scheduled tasks, delegating work to another agent, using browser or sandbox capabilities, sending messages to Feishu/channels, or viewing/sending to sessions. Requires Gateway connection.

Structured knowledge (vacuole): search, learn, and evolve. Use when (1) answering from docs/FAQ/knowledge-base — call knowledge_search (optional topic) then knowledge_get; (2) user says 'learn this' or 'remember this' — knowledge_learn(text, topic?, point?) or knowledge_add; (3) 'learn from the web' — web_search then knowledge_learn_from_urls; (4) list topics or points — knowledge_list_topics, knowledge_list_points(topic); (5) turn a topic into a dedicated skill — knowledge_skill_create(topic); (6) after bulk edits to KNOWLEDGE.md or knowledge/* — knowledge_sync. Workspace is KNOWLEDGE.md and knowledge/*.md under agent dir; base_skill does raw read/write; this skill adds search, topic/point layout, and FTS5 index.

Long-term memory (vacuole): search, read, store, forget, sync. Use when (1) user asks about past decisions, preferences, people, todos, or dates — call memory_search or memory_recall then memory_get; (2) user says 'remember this' — memory_store(text, target?); (3) mark something to forget — memory_forget(text); (4) after bulk edits to MEMORY.md or memory/* — memory_sync. Workspace: MEMORY.md and memory/YYYY-MM-DD.md under agent dir (default ./.u or MEMORY_WORKSPACE). base_skill does raw read/write; this skill adds search (FTS5, optional hybrid embedding), Store/Forgotten sections, and daily logs.

Create, wrap, or start an L3 node and connect it to the Gateway. A node wraps any software or service (browser, sandbox, local tools, APIs, connectors) into a Gateway capability: it connects with role=node, declares capabilities[], and handles node.invoke.request — analogous to an MCP server. Use when the user wants to add browser capability (start browser-node), run sandbox commands (start sandbox-node), wrap a new service as a custom node, or understand how to expose any tool to agents via gateway_node_invoke.

In-session todo list for task breakdown and progress. Use when (1) breaking a complex task into steps — todowrite(items) with status pending; (2) starting an item — todowrite(merge: true, items: [{ id, status: in_progress }]); (3) marking done — todowrite(merge: true, items: [{ id, status: completed }]); (4) checking progress — todoread. Stored in .monou/todos.json per workspace. Aligns with OpenCode-style task management.

Minimal terminal coding harness. Use when the user wants coding-agent behavior: read files, run bash commands, edit files with exact replace, write new files; prefer read-before-edit, grep/find/ls for discovery; concise responses with clear paths.

代码库发现与导航。提供 grep（内容搜索）、glob（文件名匹配）、list（目录列表），与 base_skill 的 read/edit/write 配合使用。

Create and connect new Paramecium agents. Use when the user wants to spawn a new agent (clone from template, new AGENT_ID), run it locally or prepare for remote deploy. Provides scripts and references for creating agent dirs and connecting to Gateway.

Discover Browser Node(s), open URLs (browser_fetch), get links (browser_links), click elements (browser_click), fill inputs (browser_fill). Screenshots render as images in chat.

会话内待办列表。复杂任务先拆成 todowrite，执行中用 todoread 查看、用 todowrite 更新状态（in_progress/completed），与 OpenCode 任务管理一致。

委托到其他 Agent：查在线 agent 列表、向指定 agent 的 session 发消息，由对方在该 session 内直接回复。

委托到其他 Agent：查在线 agent 列表、向指定 agent 的 session 发消息，由对方在该 session 内直接回复。

工作区知识库的检索、自学习（topic/知识点 + 从 URL 抓取）与知识转 Skill。topic 为文件夹，其下为知识点（如 股票/K线、高中数学/函数）；可 web_search + knowledge_learn_from_urls 自主学习。

Fetch URL content or search the web. Use web_fetch for a known URL; use web_search when you need to find current information.

List sessions, preview session info, or send a message to another session. Requires Gateway.