kilocode-agents | Prompt and agent presets for stronger multi-agent workflows in Kilocode

[galpt]

I built kilocode-agents to help with shipping, and we’re actually using it internally to push code to production.

It’s designed for kilocode v7.2.4 or newer and is built with ease of use in mind. Just a quick heads-up: it's worth reading through the README to make sure everything is set up correctly.

I think this project might help address some of the points in these discussions (especially #8152):

1. 【Architect Mode 】Please bring back Architect Mode in v7.x, Plan mode cannot replace architecture design #8152
2. Please release v7.x as a new separate plugin, not an overwrite update—don’t force upgrade #8155

Personally, I don't mind using v7.x as it's still very much usable. My take is that as long as we keep reporting issues on GitHub, v7.x remains a solid choice—it’s just naturally a bit incomplete compared to the stable release, which is to be expected.

This kilocode-agents is my attempt to make v7.x more engaging by creating a smarter agent pipeline. The goal is to stop switching between agent modes; instead, a single agent helps you choose the best approach for your codebase. And don’t worry, it’s not built for the hype—so you won't find any fancy buzzwords like "cognitive engine" here.

Please feel free to report any issues if you run into them!

[TzuChin]

That's fantastic. I was also just starting to work on adjusting these settings, and I was struggling with how to modify the permissions for these sub-agents from the interface, as it doesn't seem to provide any options for that. It wasn't until I carefully examined your project that I found inspiration; I had never considered configuring the settings file and then importing it.

I was also frustrated that simply using prompts didn't seem to effectively control the calling of specific sub-agents to perform tasks. It turns out there are special commands for that.

The concept I want to implement is very similar to what you have already completed, with the only difference being that I have broken down the functional roles into more detail. Thank you for sharing; I feel like I have a chance to take a big step forward.

Additionally, I'd like to ask a quick question. I took a look at the CEO agent, and the prompt seems quite extensive. Does this potentially lead to higher token consumption?

[galpt]

Hey @TzuChin, glad to hear you are finding the project helpful!

To answer your question — yes, longer prompts do consume more tokens, but it depends on how you measure. AI providers count tokens differently (OpenAI vs Anthropic for example), so I would recommend using a tokenizer like https://platform.openai.com/tokenizer to get a rough estimate for your setup.

That said, I am on the MiniMax $20 Token Plan myself, so token usage is not really a concern for me. The agents also have pre-defined Max Steps to prevent runaway loops, and I have stress-tested them on some pretty complex PRs — they hold up well.

Hope that helps!