TeleFolders: Enhanced Telegram Folders Manager Tool
Preface
I'm in a lot of groups.
So, I need to manage the chat folders.
There's already a project on GitHub: Noradrenalin-team/TeleFolders
But this project only implements basic functionality, and it's still a bit inconvenient to use.
Developed with Agent
Hermes connected to longcat-2.0
* This batch of free tokens from mimo is gone, so the only thing left on the roadside is longcat
longcat-2.0 is about one tier weaker than mimo-v2.5 (pro), so the whole process wasn't that smooth. I'm not going to share the spell (prompt).
Development Results
The interface supports English
(The original project is in Russian)
Implemented the function to create chat folders
(The original project only had an entry point for this feature)
Set filters for archived, personal, bot, group, channel
(The original project only had show/hide for archived)
CSV export / import
Github
========
Postscript
My hands-on experience with longcat-2.0 is a bit like the mimo-v2 (not v2.5) that hermes previously offered for free.
longcat-2.0 doesn't perform well with one-shot wishes. I recommend that you first ask it to come up with a design plan, then you review this plan, optimize or reject it, or have a smarter model come up with the plan. Then let it implement.
After you've discussed multiple plans with the agent, have the agent summarize the plan you've decided to use into a .md file, then you take a quick look.
Then have the agent implement according to the .md file.
This avoids the agent not understanding which plan you've ultimately chosen, and also avoids attention drift during long implementations, where it might go off and try other plans that were discussed.
* This depends on which model you're connecting to. mimo-v2.5 (pro) performs better and isn't this troublesome — you can direct it with natural language, and it understands which plan you've chosen.
longcat-2.0 has a chance of breaking the correct result of Task B while working on Task A.
So, after verifying that the functionality meets the requirements, commit to GitHub. (I know you can do local git, but for beginners I won't make it that complicated. And what if some agent deletes the entire local directory?)
If you suddenly find, oh no, I'm working on Task A, but how did Task B's effect disappear or break? You can find the commit where Task B was correct/normal, and explicitly tell the agent to analyze the code related to Task B in that commit, then compare it with the code in the current project.
For some tasks that aren't closely related to the main development line, you can get the correct answer in another way and just tell the agent directly.
Don't get tangled up in the session context trying to get it to figure out the right answer.
For example,
When longcat-2.0 said it couldn't find the official documentation for the updateDialogFilter function, I used the web version of a free claude account to find it, and directly brought the answer over.
Methods themselves:https://core.telegram.org/method/messages.updateDialogFilterRelated DialogFilter type definition:https://core.telegram.org/type/DialogFilterDetailed documentation for the "Folders" feature (recommended to focus on this, it explains the specific usage and business logic of this method):https://core.telegram.org/api/folders
It's like you're mentoring an intern. In some scenarios, letting them struggle through it becomes part of their project experience. In other scenarios, you should interrupt their struggle and just tell them the correct answer directly.
Posts
