QUICK START
Your team is live. Here's how to work with it effectively.
How to talk to your team
Your team lives in your channel. The orchestrator is your main point of contact — everything routes through it. Direct messages work, but mentioning it in the shared channel is more powerful, because every agent can see the context.
Example
@orchestrator I need a landing page for our new product. Here's the brief: [paste brief]
The orchestrator assesses the task, assigns it to the right specialist (builder for code, researcher for research), and coordinates the work. You'll see short acknowledgements and final summaries in the main channel — full logs in the specialist channels.
The channels
Main channel
Task assignments, one-line acks, and final summaries. This is where you read the headlines. Walls of text here are a bug, not a feature.
Builder channel
Full build logs, code output, errors, deployment results. Check here when you want the details.
Researcher channel
Research notes, source lists, intermediate findings. Full output here, verdict summary in the main channel.
Human-in-the-loop approvals
Some actions need your explicit approval before they proceed — deploying to production, anything with billing or real stakes. The team posts an approval gate to your main channel for these.
Approve it and it proceeds. Don't, and it waits. The team can't cross a gate without you.
The task board
A few times a day, the orchestrator posts a board showing all open work — who's on what, what's blocked, and what needs your input.
The "waiting on you" section is mandatory. If something is blocked on you, it's listed there with exactly what you need to do.
Memory and context
The team keeps memory, but agents reset between sessions. That's by design. The system periodically harvests channel history and writes structured notes, and each agent reads them when it wakes up.
Which means: say important things in the channel. Decisions, preferences, constraints — say them out loud. Don't assume the team will remember something from a private aside.
Task tracking
If you connected a task tracker (e.g. a GitHub org) during setup, the team can create tickets, claim them, do the work, and open pull requests. The flow:
- You or the orchestrator creates a ticket
- The board surfaces it as unclaimed
- The builder claims it and acknowledges in the channel
- The builder works, pushes, opens a pull request
- You review and merge — or send it back for fixes
You can also just describe a task in the channel — the orchestrator decides whether to file a ticket or handle it directly.
Runs itself
Security
Regular secret and dependency scanning across your repos. Failures escalate to the main channel.
Backups
The team's memory, protocols, and docs are backed up regularly. Your context is never lost.
Memory harvest
Channel history harvested and summarized so each session starts with current context.
Status board
Work status posted on a schedule — blocked items, unclaimed tickets, and anything waiting on you.
When things go wrong
Agents can get stuck, confused, or blocked. The right response:
- Check the specialist's channel — the full log is there
- If it's blocked on an approval, approve it
- If it's confused about context, add more in the channel (it becomes memory)
- If it's genuinely broken, email otto@imago.eco and we'll help