How to keep the model/the cli agent working on a task list

[madflow]

I have created a task list like so:

# Agent Tasks

## Goal

Build a CLI tool that ...

## Task 1: Create a README file
Document the purpose of the program...
- [] Project description and goals

## Task 2: Create a LICENSE file
- [] Create a LICENSE file with the MIT License

## Task 3: Clean up Cobra CLI boilerplate
- [] Remove the toggle option from the root command
- [] Adjust the root command to resemble this project's structure and purpose
- [] Remove unnecessary comments and placeholder text
- [] Update command descriptions to match the agent context tool functionality

## Task 4: Create internal folder and restructure project
- [] Create a folder called "internal" for all code
- [] Move the cmd folder to the internal folder

## Task 5: Implement root command with git repository detection
- [] Create the root command logic that runs when CLI is executed without arguments
- [] Implement a check to detect if we are in an existing codebase

My instruction was:

Implment every task in tasks.md. when you have completed a task, run go mod tidy and go run . to check if everything is working. when it is working, check of the tasks, and go to the next one. in no circumstance, you are allowed to stop implementing the tasks, unless they are all done and the checks are working.

I used Qwen: Qwen3 Coder (free) via Openrouter.

I had to instruct the modal 6-7 times to "continue".

Is this something which has to do with the model, the provider, the Cli tool (crush) ?

Does anyone has any tips to make this work with "crush" ?

[madflow]

In opencode, these are usually "model issues", have not checked back with this cli.