Agents

An Agent is a “dedicated colleague” with a persona, a scope of capabilities, and use cases. Compared with a general conversation, an Agent bakes in the way a particular role or scenario gets things done — hand a task to the right Agent and it completes the work in a more professional, more consistent way.

1 · Overview

An Agent typically includes: its role definition (Persona), the scenarios it excels at, the range of Skills it can use, and the boundaries it is authorized within the organization. You can bind an Agent to a conversation, a channel, or a scheduled task so it consistently handles a certain kind of work.

2 · Page overview

The Agents page has two parts:

〔Figure: the Agents page, highlighting “My agents” and the Agent store〕

• My agents — Agents you have imported/obtained and can use directly.
• Agent store — browse and obtain more Agents (some capabilities are opening up gradually).

Open an Agent to view: its Persona summary, use cases, capabilities and permissions, the Skills it includes, and its organization authorization scope.

3 · Getting and importing

You can import a local Agent:

1. Import a folder or import a ZIP (must contain CLAUDE.md or AGENTS.md).
2. Security review — before import, Aureink runs a Security Review / AI analysis to assess its capabilities and risk level (low / medium / high), and shows the file structure and included Skills.
3. Edit metadata — confirm the name, description, and available Skill scope (all Skills / only the Skills in the pack).

〔Figure: the security review and risk level shown during import〕

4 · Creating an Agent

Don’t want to write one from scratch? Use the Agent Creator to generate your own Agent step by step through a conversational interview: answer a few questions about the role, responsibilities, and scenarios, and Aureink shapes it for you.

5 · Updating and publishing

• Binding for use — choose an Agent for a conversation when creating a task (it cannot be changed after the first message is sent); channels and scheduled tasks can bind an Agent too.
• Update management — check for updates / update now; the update policy can be set to manual / auto minor version / locked.
• Publish to the store — publish your own work as an Agent or Skill, with a version number and release notes, for reuse within the organization (subject to review).

6 · Important notes

Important notes

• Permission boundaries: the Skills an Agent can use and the operations it can perform are jointly constrained by its capability scope, the permission mode, and the sandbox.
• Security review: before importing an Agent of unknown origin, review the security review results and risk level.
• Organization authorization: in organization mode, an Agent’s available scope follows Console’s grants — the client cannot escalate its own privileges.
• Quota reminder: an Agent consumes usage quota when it performs tasks.

Disclaimer: an Agent may contain third-party logic and Skills, and its behavior is determined by its own configuration. Use it only after you understand what it is for.

Last updated · July 2026 ｜ Next: Memory →