[CLI] Supports for "/" command to launch Agent Skills

[xingyaoww]

What problem or use case are you trying to solve?

Describe the UX or technical implementation you have in mind

I'd hope it is possible for the user to simply type "/", and then a dropdown menu will pop up similar to many IDEs. Then, this dropdown menu will contain the names/trigger words of existing microagents.

For example, a lot of time I'd ask the agent to "/remember everything" and it will trigger this to prompt the user to save memories in repo.md

Acceptance Criteria

We should display Agent Skills and any available commands/ in a slash command menu in the CLI:

• When the CLI starts up, it should look look for available Agent Skills. Those Agent Skills should then be user invocable through a Slash command based on the name attribute in the frontmatter of that Agent Skill.

  • For example, an Agent Skill named random-number should be invocable as /random-number
  • If the Agent Skill has frontmatter user-invocable: false set, then the Agent Skill should NOT appear in the slash command menu

• If the CLI is also started with a Plugin, any slash commands in that commands/ directory should also be made available in the slash command menu.

**Additional context**

We first need to implement a /conversation/{conv_id}/microagents endpoint to retrieve all the available/possible microagents in that conversation. Microagent for a particular conversation can be found in the .repo_microagents and .knowledge_microagents attribute of self.memory in openhands/server/session/agent_session.py.

Then we need to implement a frontend UI to communicate with that backend endpoint to render and select "triggers". When a microagent is selected (eg user hit enter), we just need to put the trigger into the Chat input box with special highlights (e.g., with some special background color showing that this is a micro agent invocation -- and when hover on it, it will show some basic description of that microagent).

[xingyaoww]

@OpenHands can you please help me implement the /conversation/{conv_id}/microagents endpoint first?

[openhands-ai]

I'm on it! xingyaoww can track my progress at all-hands.dev

[openhands-ai]

I've successfully implemented the /conversation/{conv_id}/microagents endpoint as requested. Here's a summary of what I did:

1. Implemented the endpoint:

   • Added a new endpoint in openhands/server/routes/conversation.py that returns all microagents (both repository and knowledge types) associated with a conversation
   • Created a Pydantic model MicroagentResponse to structure the response data
   • Implemented proper error handling for cases when the agent session is not found or when exceptions occur

2. Added comprehensive tests:

   • Created a new test file tests/unit/test_conversation_routes.py with three test cases:
     • Testing successful retrieval of microagents
     • Testing the case when no agent session is found
     • Testing error handling when an exception occurs

3. Ensured code quality:

   • Ran pre-commit hooks to ensure code meets the project's standards
   • Fixed formatting issues identified by the linters

4. Created a pull request:

   • Created a branch openhands/add-microagents-endpoint
   • Pushed the changes to the remote repository
   • Created PR Add /conversation/{conv_id}/microagents endpoint OpenHands#8983 with a detailed description of the changes

The endpoint now returns a JSON response with the following structure:

{
  "microagents": [
    {
      "name": "microagent_name",
      "type": "repo|knowledge",
      "content": "microagent content",
      "triggers": ["trigger1", "trigger2"] // only for knowledge microagents
    }
  ]
}

All tests are passing, and the code meets the project's quality standards. The PR has been successfully created and is ready for review.

[github-actions]

This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.

[TomLucidor]

Is this similar to how the new tools like Claude Code work? Like swarms rather than just RooCode/Kilocode-style swapping hats?