Permissions, memory, and background tasks

Agent can perform file, terminal, and network operations, and can also call sub-agents and background tasks. Permissions determine whether it should ask first, memory determines whether it will still know next time, and the status panel on the right tells you what it is doing now.

Daily use starts with default permissions; plan-only is suitable when you want to review the approach first, and full access should only be used for tasks with clear boundaries and recoverable changes.

Permission Modes

Mode	Behavior	Applicable scenario	Notes
[Confirm each time]	Ask before editing files or running commands	Default starting point, real project directory	Frequent confirmations when there are many operations, but easiest to control
【Auto-accept edits】	Can edit files, ask before running commands	Document organization, recoverable code changes	Confirm the working directory and version control status first
【Smart approval】	Whether to allow through is decided by safety checks	Supported by the model and clear task boundaries	Some models may still ask one by one
【Plan only】	Plan only, do not edit files	Proposal review, pre-launch review	Suitable for reviewing the plan before executing it
【Full access】	Skip permission checks	An isolated, trusted, recoverable environment	May delete files or access the network; highest risk

Configuration path: left navigation【Work】→ Agent menu→【Edit】→【Basic】→【Permission mode】.

Channels can override the Agent permission mode separately. External chat entry points are more likely to receive unexpected instructions, so you should usually choose【Inherit agent settings】or a stricter mode than the Agent; do not open full access just to reduce the number of confirmations.

Agent memory

Agent memory follows the Agent, not a specific task or working directory. It is suitable for storing long-lived preferences, project facts, technical decisions, and experience; one-off progress should be recorded as time-stamped logs for later retrieval by subsequent tasks.

Tell the Agent directly in【Work】:

Remember: all Chinese copy released externally should use full-width punctuation and not use exaggerated headlines. Follow this rule for related tasks in the future.

When correction is needed, clearly state that the old information is no longer valid and ask to update the memory. Do not write passwords, API keys, private identity information, or short-term irrelevant content into long-term memory.

Sub-agents, workflows, and background commands

For complex tasks, the Agent can assign research, organization, and verification to sub-agents, or orchestrate multiple steps through workflows. Time-consuming commands can run in the background without blocking the entire conversation.

In the right-side【Status】you can check:

• Tasks in progress and completed tasks;
• Sub-agents and workflows;
• Background commands and stop entry;
• Tool call successes, failures, and counts;
• Context usage and declared artifacts.

Long tasks can view artifacts, subtasks, background commands, and context usage in【Status】.

User case: continuously maintaining project standards

The team wrote stable coding conventions into Agent memory and made detailed review steps into skills. Each time a new branch is reviewed, start a task and have the Agent use sub-agents to check the interface, data migration, and tests separately, then have the main Agent summarize the conclusions. When the rules change, update the memory; there is no need to modify the historical records of every task.

Can memory replace a knowledge base?

No. Memory is suitable for a small number of stable facts and experiences that are reused across tasks; a knowledge base is suitable for systematic documentation and can control the scope of the Agent's retrieval.

If the task seems stuck, where should you look first?

First open【Status】on the right and check whether it is waiting for permission, whether a background command is still running, whether a subtask failed, or whether the provider is retrying. If you need deeper request details, open developer mode to view the call chain.