# Usage Notes

## Meeting mode

Use this mode when multiple roles should discuss a topic.

Suggested inputs:
- topic
- participant roles
- optional background context
- max rounds

Human-in-the-loop points:
- before first round
- after each role response
- after final report

Suggested user flow:
1. Describe the topic.
2. Let AI suggest roles.
3. Review the suggested roles.
4. Adjust participants or context.
5. Start the meeting.
6. Intervene if the discussion goes sideways.

Suggested default roles:
- `real_powell`
- `gstack_commercial`
- `gstack_technical`
- `gstack_ux`

## Large topic or background text

Yes — this is considered.

Recommended ways to provide large text:

### 1. Paste the full text into the context field

Best when the text is not too huge and you want the full detail available immediately.

### 2. Use a file or document as the source of truth

Best when the background is long, structured, or likely to change.

Typical pattern:
- store the full background in a document
- paste or load a short summary into the meeting prompt
- keep the full text available for reference if needed

### 3. Split the background into sections

Best when the material is large and you want the meeting to focus cleanly.

For example:
- problem statement
- current state
- constraints
- options
- risks
- decision criteria

### 4. Ask AI to compress the background first

Best when you have a long blob of text and want a cleaner pre-meeting brief.

The workflow should support:
- raw background input
- summarized background
- optional follow-up questions before the meeting starts

## Message style

Yes — each speaker should appear as a separate, standalone message.

That means:
- one message for each role turn
- one message for PM routing summaries if you want them visible
- one message for user feedback or intervention
- one message for the final report

This keeps the discussion easier to scan and makes the flow feel clean.

## Single-role mode

Use this mode when one role should judge a topic directly.

Suggested inputs:
- role name
- topic
- optional context

Suggested user flow:
1. Pick a role like `real_powell`.
2. Provide the topic.
3. Optionally add background context.
4. Read the role's judgment.
5. Ask for a second pass if needed.

## Startup brief

Before launching a meeting, the workflow should show a short brief:
- the topic
- AI-suggested roles
- context summary
- max rounds
- a note telling the user they can adjust before execution

## Behavior

- Roles are loaded from `/root/projects/meeting/agents/*.json`
- The meeting state is stored in a shared scratchpad
- The PM router decides whether to call another role or finish
- The workflow stays inside OpenClaw instead of shelling out to the meeting CLI