# Meeting Orchestrator User Manual

## What this skill is for

This skill helps you run a meeting with multiple AI roles or ask one role to judge a topic directly.

Use it when you want to:
- compare several viewpoints
- pressure-test an idea
- evaluate a strategy, product, or technical plan
- ask `real_powell` or another role for a focused opinion

## Two modes

### 1. Meeting mode

Use this when you want multiple roles to discuss one topic.

Typical questions:
- What roles should participate?
- What background should they see?
- How many rounds should the meeting run?
- What should the final report answer?

### 2. Single-role mode

Use this when you want one role to speak directly.

Typical questions:
- What does `real_powell` think?
- How would `gstack_technical` judge this architecture?
- What would `gstack_ux` say about this flow?

## Before you start

Prepare these inputs:
- topic
- optional background text
- optional files or documents
- optional role preferences
- optional round limit

If your background text is long, do not panic.
The skill is designed to handle large context in a staged way.

## How to provide a large background

You can use any of these:
- paste the full text
- give a document or file source
- break it into sections
- ask the AI to summarize it first

Recommended structure for large context:
1. problem statement
2. current situation
3. constraints
4. options
5. risks
6. decision criteria

## How meeting mode works

1. You give the topic and background.
2. AI suggests the best participants.
3. You review and confirm the role list.
4. The skill shows a startup brief.
5. The meeting begins.
6. Each role speaks in one standalone message.
7. You can interrupt, add context, or steer the meeting.
8. The meeting ends with a report.

## How role selection works

The skill prefers roles that fit the topic.
A common default set is:
- `real_powell`
- `gstack_commercial`
- `gstack_technical`
- `gstack_ux`

The AI may recommend other roles if the topic calls for them.
You should always review the suggestion before starting the meeting.

## How message output works

The discussion is intentionally clean:
- one role = one message
- one interruption = one message
- one final report = one message

This makes the meeting easier to scan and easier to continue later.

## What you can do during a meeting

You can:
- add more context
- remove a role
- add a role
- ask for a second pass
- continue the meeting after the first report

## Best practices

- Give the topic clearly.
- Provide enough background, but not a random dump with no structure.
- Let AI suggest the first participant list.
- Confirm the roles before starting.
- Keep each role turn separate.
- Use `real_powell` when you want hard judgment and synthesis.

## Example workflows

### Product review

Topic: Should we launch this feature now?
Suggested roles:
- `gstack_commercial`
- `gstack_technical`
- `gstack_ux`
- `real_powell`

### Strategy review

Topic: Should the team bet on this market direction?
Suggested roles:
- `real_powell`
- `gstack_bezos`
- `gstack_munger`
- `gstack_grove`

### Technical review

Topic: Is this architecture worth shipping?
Suggested roles:
- `ecc_architect`
- `ecc_performance_optimizer`
- `ecc_security_reviewer`
- `real_powell`

## Common mistakes

- Putting too little context into the meeting
- Skipping human confirmation of roles
- Mixing too many unrelated issues into one meeting
- Expecting the meeting to solve a vague problem without a clear topic

## Short version

If you remember only one thing:
- give the topic
- give the background
- let AI suggest roles
- confirm roles
- let the meeting run one speaker at a time