Getting claude to use agents without explicitly telling it to.

[chiefMarlin]

Is anyone else having issues with claude cli not using any of the agents unless you tell it to, for every step of the way ?
For every message i send, unless if i tell it to use a specific agent, it will just use the general context without an agent.

[ehansonddh]

Yeah, you need to be explicit or it won't do so. One way I've gotten around that - outside of using the workflows in https://github.com/wshobson/commands - is to have the prompt-agent write the prompt and tell it that it needs to specify which agents to use on which tasks.

[chiefMarlin]

@ehansonddh Thanks, could you share an example prompt that you use for prompt-agent ?

[ehansonddh]

Sure:

Have the prompt-engineer create a prompt, specifying the agents to use in each task, that creates a deployment document for the data lakehouse. This document should be written from the perspective of creating a production environment and include the following:

1. Instructions for how to set up the environment on a server running Ubuntu 22.04.
2. Instructions for environment variables:
   • Where to add each environment variable.
   • What each environment variable is for.
3. Instructions for how to initialize the DuckDB catalog on the local file system.
4. Instructions for how to set up PostgreSQL.
5. Instructions for how to set up Redis.
6. Instructions for how to set up the API.
7. Instructions for how to secure the installation after deploying it.
8. Instructions for how to monitor the health of the system after deploying it.
9. A post-deployment checklist to run after completing all of the other remaining steps.