Anyone else having trouble getting Goose to do anything useful (autonomously)?

[Stitch10925]

For days I have been trying different things with Goose. I started with trying to get a development workflow going using various recipes (project manager, tech lead, developer, tester, technical writer). Tuning, optimizing, asking other AI tools to help me optimize the recipes... yet, I haven't been able to get anything working or useful out of it.

So instead I decided to try something smaller. A research task. So I go into /plan mode, tell it what I want to do. The first time it creates a lovely plan, then it asks me if I want to execute it. Well, yes. Then it just jumps to the prompt. When I ask it to make a plan.md file for future reference, it doesn't. I need to continually babysit it and tell it to "continue" or "resume". Headless mode or not, it asks feedback.

I was able to get it to complete my research task by asking it to continue about 5 times. Still need to verify the result.

I've tried different models and context sizes, but as of now, no luck. It often happens that when I ask something, it jumps to the prompt without doing anything.

Does anyone have this issue, and ideally, managed to overcome it?

Specs:

• NVidia 16GB VRAM
• 32 GB RAM
• OpenWebUI backed by Ollama
• Tested models: qwen3:14b, gemma4:e4b, qwen3.5:9b