Glossary
Grouped by theme, with the chapter that covers each in detail. Chinese screen labels are given in brackets so you can match them on a Chinese interface.
Tasks and the window
Workspace (工作空间) The folder a task mainly reads from and writes to — whichever directory you chose when creating the task, or one WorkBuddy created. It is also what default permissions use to judge whether an operation is within scope. → 05 The new-task bar
Task One conversation session. Each is independent, with its own workspace and context. Parallel tasks are supported.
Artifact (产物) A file newly generated in the current conversation. Distinct from workspace files: artifacts are what this run produced, workspace files are everything in the directory. → 04 The three-pane window
Changes (变更) Every modification this task made to files, with a diff view. The last self-check before you accept a result. → 04 The three-pane window
Work mode Three scene entry points — daily office (日常办公), code development (代码开发), design ideas (设计创意) — with different defaults and output habits. → 01 What WorkBuddy is
Agent / Plan / Ask Three execution styles. Agent (default) executes directly; Plan produces a reviewable plan and waits; Ask answers without executing. → 06 Work modes and model tiers
Chip A capsule of content in the input box. Pastes over 3,000 characters compress into one, and chat logs forwarded from WeChat also arrive as chips. → 04 The three-pane window
Extending capability
Skill (技能) A package of executable scripts and workflows that lets the AI perform concrete actions with your authorisation. A Skill is capability. → 01 Skills
Connector (连接器) A bridge between WorkBuddy and an external service. Comes in MCP + CLI and Skill + CLI forms. → 02 Connectors
MCP (Model Context Protocol) The standard way of wiring external tools and services into WorkBuddy. The official analogy is "USB for AI". → 03 MCP and plugins
Plugin (插件) A bundle of related capabilities: Skills, MCP connections, slash commands, hooks, agents. → 03 MCP and plugins
Expert (专家) A role-switching mechanism — persona plus methodology plus toolchain, so WorkBuddy executes as a specialist. An expert is capability plus experience. → 04 Experts and expert teams
Expert team (专家团) A collaboration mechanism — several experts dividing work, with a lead that decomposes, parallelises and integrates. An expert team is several experts plus a process. Costs typically 3–5× a single expert. → 04 Experts and expert teams
Inspiration (灵感) A gallery of finished work built with these capabilities, with a one-click "make one like this" that pre-fills the prompt and loads the associated Skills and experts.
Slash command (快捷指令) A short command in the input box, e.g. /weekly. Usually arrives with a plugin.
Collaboration and accumulation
Project (项目) The container for team collaboration, organising instructions, connectors, experts, Skills and material so they inject automatically into member tasks. → 05 Projects and teamwork
Project instructions Global behaviour rules for the AI, inherited by every task in the project. The foundation of team collaboration. → 05 Projects and teamwork
Public / personal authorisation The two connector authorisation modes in a project. Public is configured once by an admin and shared; personal is each member's own account. Collaborative tasks disable personal-authorisation connectors entirely. → 05 Projects and teamwork
Task hand-off (任务流转) Moving a task's output and context to someone else. The system packages artifacts, a progress summary, custom fields and attachments. → 05 Projects and teamwork
Library (资料库) Where humans and agents both keep output, split into "My documents" (personal) and "Team space" (shared). → 06 The Library
MD / CSV / HTML split The Library's core design: MD decides content (narrative), CSV stores data (single source of truth), HTML handles expression (presentation). Change content in MD, data in CSV, appearance in HTML. → 06 The Library
Publish as website (发布为网站) Turning Library HTML into a public link under workbuddy.link. Off by default; once on, anyone with the link can open it. → 06 The Library
Memory (记忆) Personal information extracted from conversation history (facts, preferences, relationships, follow-ups), updated nightly, visible only to you. Costs no credits. → 09 Memory
Automation and remote work
Automation (自动化) A task triggered on a schedule. Configuration lives on the local client and runs as your signed-in identity. Personal — visible only to its creator. → 07 Automations on a schedule
Assistant (助理) Remote control: send tasks to your desktop WorkBuddy from WeChat, WeCom, QQ, Feishu or DingTalk. → 08 Assistant
Assistant folder The fixed working directory assistant tasks use, not changeable. One of the key differences from an ordinary task.
Permissions and security
Default permissions (默认权限) The recommended everyday mode: efficient inside the workspace, stopping to ask before high-risk actions. Scoped to the current task. → 10 Permission modes and the sandbox
Full access (完全访问权限) Turns off second confirmation. A global switch — it covers every task in this client, including ones created afterwards. → 10 Permission modes and the sandbox
Sandbox The isolated environment commands run in. Commands run under sandbox constraints first; if blocked, risk determines whether you get asked.
Security Centre (安全中心) Where security and runtime configuration lives: program blocklist, file read/write rules, automatic backup, malicious-domain blocking. → 10 Permission modes and the sandbox
Program blocklist Programs the agent and its subprocesses cannot launch. Blocked outright, with no prompt.
Forced-approval blocklist Interception rules for sensitive directories, in "write" and "read/write" tiers. Default rules already include read protection for directories such as ssh.
Automatic backup A snapshot taken before the agent modifies an existing file. Within one conversation turn, a file gets one snapshot before its first modification. All sessions share the space, capped at 3000 MB by default.
Models and cost
Fast / balanced / maximum The three model tiers. Fast is cheapest and quickest, maximum is most reliable and priciest, balanced is the default recommendation. → 06 Work modes and model tiers
Thinking effort How much reasoning the model invests before answering. Higher means more analysis and self-checking, and slower responses. The default suits most tasks.
Context window 200K by default. Every turn sends the full current context to the model, so a bigger window costs more per turn. Going to 1M does not make it smarter.
Max mode Off, the system compresses context automatically; on, it is preserved in full and long conversations burn credits noticeably faster. Leave it off.
Credits (积分) The unit metering AI resource consumption. Every conversation-related request costs credits, scaled by model pricing and task complexity. Issued monthly, valid that month, no roll-over. → Credits & plans
Top-up pack Supplementary credits when the monthly allowance runs short. Unaffected by an upgrade — validity and balance are preserved.
Developer capabilities
Worktree task Uses Git worktrees to give each parallel task a fully isolated working directory while sharing branches and objects with the main repository. → 11 Git and parallel worktrees
Base branch The branch a worktree task is created from. Locked after creation.
Task branch The branch a worktree task creates automatically, named workbuddy/{base}-{8-char hash}. The prefix distinguishes it from branches you made.
Enterprise Agent A cloud agent runtime platform with lifecycle management: runtimes (cloud sandbox instances), sessions, manifests (declarative configuration), versions and snapshots, deployment. Enterprise plans only.
Easily confused pairs
| This | That | The difference |
|---|---|---|
| Skill | Expert | A Skill is capability (can do something); an expert is capability plus experience (understands a field) |
| Skill | Connector | A connector solves external-service integration with a standard authorisation flow; a Skill is a broader capability wrapper |
| Connector | MCP | A connector is a product-level concept; some official connectors are MCP underneath |
| Artifacts | Workspace files | Artifacts are what this run produced; workspace files are everything in the directory |
| Memory | Project instructions | Memory is extracted personal preference; project instructions are hand-written team convention |
| Disable a Skill | Uninstall a Skill | Disabling pauses model calls, keeping files and config; uninstalling deletes everything |
| Share a task | Hand off a task | Sharing brings someone into the same conversation; hand-off moves the output and context to them |
| Default permissions | Full access | The first is scoped to the current task; the second covers every task in the client |