Skip to content

Latest commit

 

History

History

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 
 
 

README.md

/queue

Send your agent a list of prompts after it's finished working.

Why? Because agents don't always follow guidelines put forth in skills. I've found a queue of prompts to be much more effective at enforcing rules. It's like progressively disclosing knowledge which works better for human brains, too.

Usage

/queue <key>         Run the prompt queue saved under <key>
/queue :abort        Cancel an in-progress queue
/queue :delete <key> Delete a saved queue (creates a backup snapshot)

On first run, you'll get a default queue called review-and-fix with some code review instructions. Feel free to hack this apart or whatever.

Code review example

Let's say you just gave your agent a big task. You could run /queue review-and-fix which will wait for the agent to finish and then give it a list of things to double check. One of them might be Did you use TDD? Really? I don't think so. Use the break/fix skill to validate your tests. Another could be Did you check for type errors, lint, and formatting errors? Really? I don't think so. Try again. You don't have to use it for code review. It's just a queue of prompts to do whatever.

Interactive prompt editor

Before the queue runs, you can customize each prompt, delete the ones you don't need or add new ones. Pressing enter queues all the prompts. By default, any changes will only apply to your current session but you can press S to update the saved queue.

Installation

This is a pi extension. Run /pie-kumar303-config to install it.