For AI agents: the complete documentation index is available at https://a3s-lab.github.io/UI/v0.4.0/en/llms.txt, the full documentation bundle is available at https://a3s-lab.github.io/UI/v0.4.0/en/llms-full.txt, and this page is available as Markdown at https://a3s-lab.github.io/UI/v0.4.0/en/harness/index.md.

Harness overview

Harness groups the components needed to build Coding Agent workflows. It separates task input, conversation, execution, permission, review, evidence, and preview tools from general application layout so teams can choose components by user task instead of page appearance.

Harness currently contains 30 public components. Every guide keeps its stable /components/* route, so existing links and package imports remain compatible.

Workspace framework

Dock Workspace · Grid View · Split View · Pane View

These are optional layout integrations and do not change the 116-component semantic manifest. Dock Workspace owns tabs, DnD, floating and popout groups, Edge Groups, and layout recovery. Grid, Split, and Pane View cover a fixed two-dimensional grid, a one-axis split, and collapsible tool panes. Every layout guide includes vanilla, React, Vue, and matching hooks.

Task and input

Task Start · Task Workspace · Task Composer · Context Selector

This group establishes the task, binds workspace context, and captures the first actionable instruction. Input state, available capabilities, and the submission boundary remain explicit before execution begins.

Conversation and messages

Agent Transcript · Agent Workbench · Message Status · Message Attachment · Message Citation · Follow-up Suggestions

This group presents the chronological conversation and keeps streaming, queued, offline, citation, attachment, and follow-up states attached to the message that owns them.

Files and knowledge

File Explorer · File Manager · Knowledge Library

This group manages the workspace material a task may use. File Explorer handles source hierarchy, File Manager handles familiar file operations, and Knowledge Library handles durable searchable sources with explicit indexing and recovery states.

Execution and approval

Task Plan · Plan Step · Task Queue · Approval Request · Execution Item · Checkpoint · Tool Call · Tool Call Timeline · Tool Result

This group keeps plans, execution progress, permission boundaries, and recoverable state visible. High-impact actions disclose their source, scope, and authorization result before execution.

Review and evidence

Change Review · Execution Evidence · Artifact Card

This group turns execution results into inspectable, traceable evidence and gives reviewers a clear basis for accepting, revising, or reverting the work.

Development and preview

3D Code Graph · Code Editor · Terminal · Log Viewer · Device Simulator

Development and preview components inspect and operate on real work results. Code, terminal, log, and device capabilities stay inside their sandbox boundary; operations that need local files, processes, or native views are handed to a trusted host through structured events.