several Hermes agents working together: (1) a profile or a separate install for a specialist agent? a profile. each one is an independent Hermes agent with its own config, API keys, memory, sessions and skills. give it a description when you create it, so the Kanban orchestrator knows what it is good at: ``` hermes profile create researcher --description "Reads source code and external docs, writes findings." ``` (2) a subagent or a Kanban task? a subagent from `delegate_task` starts with a fresh conversation and hands back its final summary, so a big job stays out of your context. use one when your agent needs an answer before it carries on. a Kanban task goes to a named profile with its own memory. use one when the work crosses agents or has to survive a restart. (3) how do agents hand work to each other? through the Kanban board. create a task with an assignee, and the dispatcher in your gateway starts that profile as a worker. the handoff happens in comments: a worker reads the whole thread when it starts. a task with a parent waits until the parent is done: ``` hermes kanban create "research the market" --assignee researcher hermes kanban create "write the report" --assignee writer --parent <task-id> ``` (4) can my bots talk to each other in a group chat? yes. Hermes ignores other bots by default, so you opt in per platform. with `mentions`, a bot answers another bot when that message tags it: ``` DISCORD_ALLOW_BOTS=mentions ``` behind that sits a loop guard: once 20 bot messages land in one channel inside 5 minutes, further bot messages there are dropped for 10 minutes. human messages are left out of the count. (5) how do I check what the workers did? a worker ends its task with a summary of what it did, or blocks it and gives the reason. `hermes kanban runs <id>` lists every attempt. for parallel work, `hermes kanban swarm` sets up the workers plus a verifier that waits for all of them.