So I am using agents to replicate a whole team and it is working fantastically. There are a few issues I ran into.

Channel chat: @mentions in a chat per project helps a ton to give tasks and contexts to agents. The problem is the UI - agents tend to give me structured and consistent replies based on their prompt structure rather than a short reply a team member would give. This breaks having the thread locked on a small window and also requires clicking on replies to be viewed. This creates issues especially when reviewing responses before its sent to another agent. Also tasks can run in tandem and is stuck in a tight window. When swopping multiple tasks with multiple reply windows this gets worse. My suggestion here is a simple reddit style thread with collapsible spines dependent on nested conversation. This also gives the visual benefit at glancing which part of the text thread your interested in using the spine tabs.

On the fly scheduled work: Right now agents cannot seem to call other agents. I understand this can create infinite token loops happening on the background but a simple approve button for approve agent call and or approve for this thread. Notifications will help keep track like a popup icon with the agent name and spining wheel. A way to steer and stop the agent if they hallucinate in the wrong the direction. This scheduled work opens up a lot of pipelines to work.

Protocols, guidelines, blacklist, whitelist: This is either group or global structured prompts that apply to all agents. For example I want to implement an repeatable loop. Doc > Review doc > QC > Doc handover port >> this gets downloaded by my hermes agent and turned into a presentation. The steps before the handover port is consistent and uses the spaces locations to call where the agents need to implement this skill. I can dedicate an agent to do this but 3 agents for each step - which are common steps regardless of protocol - but specialized agents who responses are predictable just need to know this simple pipeline without me assigning it manually. I created a simple index doc that all agents inherent that they can lookup the skill needed that has the protocol - which works. It just makes sense to have a hub for this and the advantage is grouping agents based on project or processes. Guidelines are modular system prompts that can help responses, blacklist and whitelist is just a collection of do's and dont's. I bring them up because it makes sense if they sit in the same area as its toplevel prompts.

Stronger kanban integration : Kanban is the lowest tech agent tracker. Dedicated kanban tracker just for agents to create visibility. Agents can use the existing kanban but it struggles to move through clickup. This also allows prompt injection per board phase. eg. QC phase can inject a qc prompt in the agents context. Allowing the agent to "change roles" while still being specialized in the pipeline. This brings up context clearing too as a heavier context forms a bit more confusion to the agents. Clearing context allows less conflicted prompt injections.

---

I created a dedicated clickup guide agent to help me implement these but the agent couldnt help - which means these features could be there but I don't have that visibility : Just mentioning it as an fyi. Hermes integration will be great and won't replace what the team created here: it's actually the strongest multi-agent workflow I have come across - if it had custom tool calling it would be the best imo. I do think it is slightly ahead of the time as everyone is still rushing that one prompt magic whereas treating it as a system creates rapid, tweakable and consistent results making it production ready.