Updated on 2026-07-29 GMT+08:00

Agent Chats

CodeArts Agent's agent mode understands developer needs and plans tasks automatically. It enables full process automation from requirement analysis to coding, boosting efficiency in end-to-end development.

Core Capabilities

The agent offers essential features for understanding requirements, planning tasks, writing code, and using tools.

  1. Complex task planning and management: Natural language can turn into well-structured action plans. The tool updates task lists instantly as work progresses, ensuring organized tracking and standardized handling of project activities.
  2. Autonomous coding: The tool autonomously generates and modifies code, ensuring consistent updates across multiple files even in complex situations.
  3. Intelligent tool calling: The agent retrieves information and performs tasks automatically. It accesses tools like code search, file reading, and shell commands to ensure the new code integrates smoothly with the project's current logic and dependencies.
  4. Interaction mechanism: The tool includes a manual confirmation step at critical stages to maintain security and control during operations.

Core Development Modes

The agent provides the following development modes, which you can flexibly select based on your needs.

  • Vibe coding: Coding while chatting, verify instantly. It is suitable for brainstorming and creative development, enabling rapid iteration within chats and dynamically capturing and clarifying potential requirements. It boosts efficient conversion from vague ideas into feasible prototypes.
  • Spec-driven development (SDD): An engineering methodology that guides system design, development, and verification based on explicit technical or business specifications. Under this mode, all development activities follow established specifications to ensure that the final deliverable is highly consistent with the specifications in terms of functionality, performance, compatibility, and other aspects. It is suitable for scenarios with high requirements for consistency and verifiability.
  1. Switch to the agent mode.

    1. Log in to CodeArts Agent IDE as instructed in Quick Start.
    2. At the bottom of the input box on the chat panel, choose Built-in > Agent to switch to the agent mode. The selected model is displayed on the right. You can select a model as required from the drop-down list.

      If the chat panel of CodeArts Agent is not displayed, click (Expand AI Sidebar) in the upper right corner of the top menu bar to open it.

    3. On the chat panel, click Vibe Coding.

  2. Explain what you need. Type your task into the box, and click or press Enter to submit your request.

    Generate a lightweight music player with a clean, minimalist design and a dark, immersive theme. Use a simple, uncluttered layout with no unnecessary elements. Incorporate a vinyl-inspired texture, and present the lyrics with generous spacing for a breathable visual experience.
    • During the response generation, you can initiate a new chat in the input box, and the new chat will be automatically added to the task queue. The task queue supports a maximum of 20 messages waiting to be sent, and any exceeding messages will no longer be accepted.
    • Different models may vary in their generated results; refer to the actual output for the final performance.

  3. Check and confirm the plan. The agent generates a to-do list based on your prompt.

    Figure 1 Execution plan

  4. Watch the process run. Once the plan is confirmed, the task begins step by step. The agent will walk you through each operation for confirmation. You can also expand the to-do list to view the progress of the current task.

    • : The task is completed.
    • : The task is in progress.
    • : The task is not started.

  5. Confirm that tasks are completed. After the task is complete, the changed files are stored in the temporary storage area.
  6. Confirm the result. Click the arrow next to the file changed to expand the file information. Perform operations as required.

    Figure 2 Result confirmation
    • Click the file name to view the generated content.
    • Move the cursor to the file, and click to accept it or to reject it.
    • If there are multiple files, you can click the accept all button to accept all files or click the reject all button to reject all files.

  1. Switch to the agent mode.

    1. Log in to CodeArts Agent IDE as instructed in Quick Start.
    2. At the bottom of the input box on the chat panel, choose Built-in > Agent to switch to the agent mode. The selected model is displayed on the right. You can select a model as required from the drop-down list.

      If the chat panel of CodeArts Agent is not displayed, click (Expand AI Sidebar) in the upper right corner of the top menu bar to open it.

    3. On the chat panel, click Spec-driven.

  2. View the operations step by step following the preset workflow. In the input box, enter the task you want to complete and click or press Enter. The agent will execute sequentially according to the following workflow: requirement specification design > implementation scheme creation > coding task planning > task execution. This ensures that each step advances efficiently and orderly.

    Generate the core logic of a music player according to the following specifications. Strictly follow the requirements with no omission of logic allowed.
    - Supports play, pause, previous track, next track, and loop playback (single/list).
    - The progress bar supports clicking to skip and dragging.
    - The volume range is 0 to 1, defaulted to 0.7 and with mute toggling.
    - Automatically updates the cover, song title, and artist information when switching songs.
    - Implemented using HTML+CSS+JS
    Figure 3 Executing a task based on a specified workflow
    • If you confirm that the content of the requirements document basically meets your expectations, click Accept All or click to accept the generated requirements document.
    • If you do not approve the currently generated requirements document, click Reject All or click to reject the generated requirements document. In this case, if the model cannot find the generated document in subsequent steps, it will generate the missing document again.
    • After confirming that the requirements document is correct, click Start Create Solution to initiate the solution generation process.

Beyond spec-driven development, CodeArts Agent can also use slash commands for code development.

  1. Switch to the agent mode.

    1. Log in to CodeArts Agent IDE as instructed in Quick Start.
    2. At the bottom of the input box on the chat panel, choose Built-in > Agent to switch to the agent mode. The selected model is displayed on the right. You can select a model as required from the drop-down list.

      If the chat panel of CodeArts Agent is not displayed, click (Expand AI Sidebar) in the upper right corner of the top menu bar to open it.

    3. On the chat panel, click Vibe Coding.

  2. Create requirement specifications.

    1. Enter a slash (/) in the input box and choose /sdd-new from the shortcut menu.
    2. Enter the task you want to complete (for example, writing a calculator app) and click or press Enter.
    3. The agent generates a requirement specification document named spec.md based on the task you entered. If the generated spec.md file does not meet your expectations, you can manually modify it or use the agent to modify it.
      Figure 4 Generating spec.md
    4. Open Project root directory /.codeartsdoer/specs/Name of the generated file. You can find the generated spec.md file.

  3. Create a technical design.

    Enter a slash (/) in the input box and choose /sdd-design from the shortcut menu. Then enter the artifact folder name (for example, calculator) and click or press Enter.

    The agent generates the design.md file based on the spec.md file obtained in the previous step. If the generated design.md file does not meet your expectations, you can manually modify it or use the agent to modify it.

    Figure 5 Generate design.md

  4. Create a coding task list.

    Enter a slash (/) in the input box and choose /sdd-tasks from the shortcut menu. Then enter the artifact folder name (for example, calculator) and click or press Enter.

    The agent generates the task list tasks.md based on spec.md and design.md. If the generated tasks.md file does not meet your expectations, you can manually modify it or use the agent to modify it.

  5. Perform coding tasks.

    Enter a slash (/) in the input box and choose /sdd-apply from the shortcut menu. Then enter the artifact folder name (for example, calculator) and click or press Enter.

    The agent performs coding tasks item by item according to tasks.md.

Specifying the Agent Running Mode

To ensure the security and flexibility of agent running, CodeArts Agent provides three running policies: sandbox running, automatic running, and manual running. You can select a mode based on your service requirements for security and automation. By default, the sandbox running mode is enabled for CodeArts Agent.

Setting the Automatic Execution Permission

If the automatic execution permission is not enabled and the agent needs to call tools, you need to manually approve this operation, such as selecting Run Once or Skip Unauthorized Tools. After automatic execution is enabled, the related operations are automatically approved without your intervention. In addition, when you manually confirm the operation, you can also select Ask Each Time or Always Allow and Enable Config to meet different requirements.

Turning on automatic execution can lead to operational risks. Assess these risks carefully first and only enable this feature in secure and trustworthy environment.

Optimizing Input

CodeArts Agent optimizes user input to help beginners ask questions more clearly and professionally, enabling users to obtain comprehensive, structured, and high-quality responses at one time. This effectively reduces repeated follow-up questions and enhances communication efficiency as well as user experience.

Unit Test (Agent)

Context

Give the agent details like specific files, folders, Git history, or terminal outputs. This helps them grasp your project setup and requirements clearly, minimizing errors and producing precise code.

For details, see Context.

MCP

Agents can access external data sources and tools through MCP servers to handle various service needs.

For details, see MCP.

Skills

A skill is a reusable module that encapsulates expertise like coding experience and common solutions. Agents can invoke these skills to dynamically load "skill packages" on demand, boosting efficiency and performance for specific tasks.

For details, see Skills.

Diff View

Once the agent finishes generating or optimizing code, check the modifications in the diff view. It highlights added and removed lines with distinct colors, making it easy for you to review changes and confirm accuracy.

The diff view is automatically enabled. For details, see Diff View.

Multi-Chat Management

In traditional development, tasks in a project must be executed one by one. Only one task can be executed at a time, resulting in low efficiency. With the multi-chat management function, the system can execute multiple tasks at the same time, improving the overall development efficiency.

For details, see Multi-Chat Management.

Checkpoints

Checkpoint is a built-in mechanism that enables CodeArts Agent to automatically take snapshots of code status and allow for rollback. The system will automatically track the editing operations when the agent is running. Before every time that the agent modifies code, the system will automatically save the current code and chat status for quick cancellation, rollback, and historical version recovery. This function is applicable to AI-assisted development scenarios that require quick trial-and-error, secure editing, and one-click rollback.

For details, see Checkpoints.

LSP

CodeArts Agent features a pre-installed local LSP function. Once enabled, the agent can leverage the diagnostic information, code definition, and symbol information provided by the language service, achieving more precise code understanding and modification.

For details, see LSP.