General feedback #2

New Issue

2026-02-16T03:19:20-05:00

yindo commented

2026-02-16 03:19:20 -05:00

Originally created by @adrianlyjak on GitHub (Aug 19, 2025).

1. navigation and selection is unintuitive

ctrl+q seems like the wrong choice of command to proceed (intuitively is "quit").
There's also no other way to cancel out or go back
If you ctrl+q out of the first or second screen before selecting, you'll get a traceback, if you ctrl+q after selecting, it

Additionally, this is using textual, having multiple screens with different types of selectors is a little confusing (also no way to go back?). Might be simpler to just have a single screen. It is however fundamentally 2 selectors, so I would probably recommend a better fit for prompt-toolkit or rich. Textual seems a little too UI for a simple feature. (dark mode toggle and screenshot save are a bit much 😂 )

2. multiselect

Might want to have workflows and llama cloud both in one go. Right now the service is single select. I think some of this structure might mess up how this tool works, since it outputs a single file. Also, what happens if that file already exists with user written content? I think write now it will just overwrite it

3. I love the logo

🌈

4. Programmatic interface

I'd love to use this in some way via llamactl. I think integrating as a cli might be difficult, but depending on it as a library might make more sense. Right now the core implementation is pretty simple, but I could see its complexity growing over time.

5. Selecting/deselecting an agent doesn't work

Steps to recreate:

select Github Copilot
deselect Github Copilot
Select claude code

Results in both:

.github/copilot-instructions.md
CLAUDE.md

6. Rules vs agent docs vs documentation folder

This is vague feedback, but one thing that I'm wondering is if it might be better to provide short rules that are indexes into a larger piece of documentation, such that the agent can refer on demand to necessary docs, with some helpful info about what that doc is (like an index). This gets really close to llms.txt, which is what the llama index rule is.

How should updates work in this case? It kind of needs a way to intelligently manage

Originally created by @adrianlyjak on GitHub (Aug 19, 2025). ### 1. navigation and selection is unintuitive - ctrl+q seems like the wrong choice of command to proceed (intuitively is "quit"). - There's also no other way to cancel out or go back - If you ctrl+q out of the first or second screen before selecting, you'll get a traceback, if you ctrl+q after selecting, it Additionally, this is using textual, having multiple screens with different types of selectors is a little confusing (also no way to go back?). Might be simpler to just have a single screen. It is however fundamentally 2 selectors, so I would probably recommend a better fit for prompt-toolkit or rich. Textual seems a little too UI for a simple feature. (dark mode toggle and screenshot save are a bit much 😂 ) ### 2. multiselect Might want to have workflows and llama cloud both in one go. Right now the service is single select. I think some of this structure might mess up how this tool works, since it outputs a single file. Also, what happens if that file already exists with user written content? I think write now it will just overwrite it ### 3. I love the logo 🌈 ### 4. Programmatic interface I'd love to use this in some way via llamactl. I think integrating as a cli might be difficult, but depending on it as a library might make more sense. Right now the core implementation is pretty simple, but I could see its complexity growing over time. ### 5. Selecting/deselecting an agent doesn't work Steps to recreate: - select Github Copilot - deselect Github Copilot - Select claude code Results in both: ``` .github/copilot-instructions.md CLAUDE.md ``` ### 6. Rules vs agent docs vs documentation folder This is vague feedback, but one thing that I'm wondering is if it might be better to provide short rules that are indexes into a larger piece of documentation, such that the agent can refer on demand to necessary docs, with some helpful info about what that doc is (like an index). This gets really close to llms.txt, which is what the llama index rule is. How should updates work in this case? It kind of needs a way to intelligently manage

yindo closed this issue

2026-02-16 03:19:20 -05:00

yindo commented

2026-02-16 03:19:20 -05:00

@AstraBert commented on GitHub (Aug 20, 2025):

Super valuable feedback @adrianlyjak, thank you sooo much! 🙏

All remarks make sense, and for what concerns "what happens if a file already exists?", in that case we append the instructions to the already-existing file

@AstraBert commented on GitHub (Aug 20, 2025): Super valuable feedback @adrianlyjak, thank you sooo much! 🙏 All remarks make sense, and for what concerns "what happens if a file already exists?", in that case we append the instructions to the already-existing file

yindo commented

2026-02-16 03:19:20 -05:00

@AstraBert commented on GitHub (Aug 20, 2025):

In #5 I address points 1, 2 and 5 :) For the programmatic interface, I'll take some more time to design it, but given the code we already have it shouldn't be too difficult to add it!

@AstraBert commented on GitHub (Aug 20, 2025): In #5 I address points 1, 2 and 5 :) For the programmatic interface, I'll take some more time to design it, but given the code we already have it shouldn't be too difficult to add it!

yindo referenced this issue

2026-02-16 03:19:22 -05:00

[PR #2] [MERGED] fix: typos #6

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: run-llama/vibe-llama#2