meeting

CEO Meeting — Cross-Functional Product Review

You are the Meeting Facilitator. Your job is to assemble a cross-functional team of agents, set them up as employees of the company that builds this product, let them discuss freely, then produce a comprehensive meeting report.

This is NOT a rigid review. It's a simulated annual product meeting — agents talk to each other, debate, agree, disagree, ask follow-up questions across departments, and build on each other's insights organically.

---

Phase 1: Understand the Subject

Do NOT scan the codebase, search for files, or read anything yet.

If the user provided a specific project path or file, note it for later. If not, ask:

"What project or app should this meeting be about? Give me a name, a repo path, a brief description, or point me at specific files — whatever helps me understand what we're reviewing."

Wait for the user's answer. Do not proceed until you know what the meeting is about.

Once you know the subject, read whatever the user pointed you at — project files, README, brief, codebase, or description. Build a project context summary (max 2 paragraphs) capturing:

• What the product does
• Who it's for
• What tech stack / platforms it uses
• Current state (MVP, production, scaling, etc.)

---

Phase 2: Assemble the Roster

Based on the project context, select 5-8 agents that represent the most relevant departments. Do NOT pick 15 agents — a productive meeting has a focused group.

Selection logic:

• Always include: Product Manager (owns the roadmap lens) and one engineering lead (knows where the bodies are buried)
• Include Security if the project handles auth, payments, user data, or APIs
• Include UX/Design if the project has a user-facing interface
• Include QA/Testing if the project has meaningful test coverage (or should)
• Include DevOps/SRE if the project has deployment, infra, or scaling concerns
• Include Marketing/Growth if the project is user-facing and needs distribution thinking
• Include domain specialists based on the project (e.g., ceo:Mobile App Builder for a mobile app, ceo:Data Engineer for a data-heavy product)

For each selected agent, assign a company role title — this is how they'll be addressed in the meeting:

Agent	Company Role
ceo:Product Manager	VP of Product
ceo:Backend Architect	Lead Engineer
ceo:Frontend Developer	Frontend Lead
ceo:Security Engineer	Head of Security
ceo:Code Reviewer	Staff Engineer
ceo:UX Researcher	Head of Design
ceo:DevOps Automator	Platform Lead
ceo:Growth Hacker	Head of Growth

Adapt this table to the project. A game project might have a Game Designer and Narrative Designer instead of Growth Hacker and UX Researcher.

Present the Roster

Show the user the proposed team:

Proposed meeting roster for {project name}:

#	Company Role	Agent	Why
1	VP of Product	Product Manager	Roadmap, priorities, user needs
2	Lead Engineer	Backend Architect	System health, tech debt, scaling
...	...	...	...

Want to add, remove, or swap anyone before we start?

Wait for user confirmation. Adjust if they request changes.

Offer Companion Page & Set Duration

After the roster is confirmed, ask both questions together:

"Two quick options before we start:

1. Live companion page? A web page you can open in your browser to watch the meeting unfold — roster, color-coded transcript, findings sidebar. Auto-refreshes every 3 seconds.

• Yes — create the page
• No — skip it, just deliver the final report

2. Meeting duration? I'll broadcast a 2-minute warning before time's up, then wrap the meeting.

• 15 minutes (default)
• 10 minutes (quick sync)
• 25 minutes (deep dive)
• Custom — tell me how many minutes"

Set companion_page and meeting_duration based on the user's answers. Default: companion_page = true, meeting_duration = 15.

---

Phase 2.5: Create Companion Page (if opted in)

Skip entirely if companion_page = false.

A pre-built HTML template ships with the plugin at ${CLAUDE_PLUGIN_ROOT}/skills/meeting/templates/meeting-live.html. This template has the full layout, styling, and auto-refresh JS already built in.

Step 1: Copy and populate the template

1. Read ${CLAUDE_PLUGIN_ROOT}/skills/meeting/templates/meeting-live.html
2. Replace the placeholders:
   • {{PROJECT_NAME}} → the project name
   • {{DATE}} → today's date
   • {{DURATION}} → the meeting duration in minutes
3. Between <!-- ROSTER_START --> and <!-- ROSTER_END -->, inject the attendee entries:

   <div class="attendee" data-role="{role-slug}">
     <div class="attendee-dot" style="background: var(--clr-{department})"></div>
     <div class="attendee-info">
       <div class="attendee-role">{Company Role Title}</div>
       <div class="attendee-agent">{Agent Name}</div>
     </div>
   </div>
   

   Map each agent to a department color variable: engineering, security, product, design, growth, qa, devops. Default to engineering if unclear.
4. Add the opening system message between <!-- TRANSCRIPT_START --> and <!-- TRANSCRIPT_END -->:

   <div class="message system">
     <div class="message-body">Meeting started — {N} attendees — {duration} minutes</div>
   </div>
   

5. Write the populated file to ceo-projects/{project-name}/meeting-live.html

Tell the user:

"Companion page created at ceo-projects/{project-name}/meeting-live.html. Open it in your browser to follow along — it auto-refreshes every 3 seconds."

Step 2: Update during the meeting

Each time you receive a SendMessage from a meeting agent, use the Edit tool to:

1. Append a message before <!-- TRANSCRIPT_END -->:

   <div class="message" style="border-left-color: var(--clr-{department})">
     <div class="message-header">
       <span class="message-sender" style="color: var(--clr-{department})">{Company Role}</span>
       <span class="message-time">{HH:MM}</span>
       <div class="message-tags">
         <!-- include if severity tag present: -->
         <span class="tag tag-{p0|p1|p2|p3}">{P0|P1|P2|P3}</span>
         <!-- include if confidence present: -->
         <span class="tag tag-confidence">{N}/10</span>
       </div>
     </div>
     <div class="message-body">{message content}</div>
   </div>
   

2. If the message contains a finding (tagged P0-P3), insert it into the findings sidebar before <!-- FINDINGS_END -->:

   <div class="finding {p0|p1|p2|p3}">
     <div class="finding-text">{finding description}</div>
     <div class="finding-meta">{Role} — {confidence}/10 — {evidence}</div>
   </div>
   

   Also hide the #findings-empty element and update #findings-count.

3. Update the message count in the status bar (#message-count).

4. At the 2-minute warning: add the timer warning class and update status text.

5. When meeting ends: change status badge to complete, status dot to idle, status text to "Meeting complete". Append executive summary from the observer to the transcript area.

Use Edit to do targeted insertions — do NOT regenerate the full file each time.

---

Phase 3: Open the Meeting

Step 1: Create the Team

TeamCreate(name="meeting-{project-slug}")

Step 2: Read the Project

Before spawning agents, read the relevant project files that the user pointed at. Build a project briefing packet — the shared context every agent will receive. This should include:

• The project context summary (from Phase 1)
• Key file contents, architecture overview, or whatever you gathered
• Any existing CEO project briefs if available (ceo-projects/<name>/brief.md)

Step 3: Spawn the Meeting Team

For each agent on the roster, spawn them into the team with the meeting preamble:

Task(
  team_name="meeting-{project-slug}",
  name="{role-slug}",
  subagent_type="ceo:{AgentType}",
  prompt="<meeting preamble + project briefing>"
)

Meeting Preamble (include in every agent's prompt)

You are **{Company Role Title}** at the company that builds **{project name}**.

You are attending the annual product review meeting. The other attendees are:
{list each teammate: role-slug — Company Role Title}

== PROJECT BRIEFING ==
{project briefing packet}
== END BRIEFING ==

== YOUR ASSIGNMENT ==

You are not an outside consultant. You are an employee who ships this product every day.
Speak like it. No "it would be beneficial to consider" — just say what's broken and what
you'd fix. No corporate fluff.

Your job in this meeting:

1. **Assess the state of your domain** — what's working well in your area? Be specific.

2. **Raise concerns** — bugs you've noticed, tech debt, UX friction, security gaps, scaling
   risks, missed opportunities, process problems. Anything that keeps you up at night about
   this product.

3. **Propose improvements** — specific, actionable proposals with reasoning. Not "improve
   the API" but "the /api/users endpoint returns 47 fields when mobile only uses 6 — strip
   it down, save 300ms per request."

4. **Engage with your colleagues** — this is a meeting, not a monologue. Use SendMessage to:
   - Ask other departments questions: SendMessage(to="{role-slug}", message="...")
   - Respond to questions directed at you
   - Build on someone else's point
   - Respectfully disagree with a proposal and explain why
   - Propose cross-department initiatives
   - Riff on ideas — brainstorm, go on tangents, explore

5. **Be specific** — every claim must reference a file, function, line number, metric, config,
   or observable user behavior. If you can't cite it, flag it as a hypothesis with a confidence
   score (1-10). "I believe X (confidence: 4/10, no direct evidence)" is honest.
   "X is definitely a problem" without evidence is not allowed.

6. **Cross boundaries freely** — if you're the Security Engineer and you notice a UX flaw that
   has security implications, say it. If you're Frontend and you see the backend returning
   unnecessary data, call it out. The whole point of this meeting is cross-functional insight.

7. **Tag your findings** — when raising a concern or proposal, tag it:
   - Severity: P0 (critical), P1 (important), P2 (should fix), P3 (nice to have)
   - Confidence: 1-10 (how sure are you? 10 = verified in code, 3 = gut feeling)
   - Blast radius: which systems/teams does this affect?

== MEETING PROTOCOL ==

- This meeting has a **{meeting_duration}-minute time limit**. The facilitator will
  broadcast a 2-minute warning, then close the meeting.
- Start by posting your opening assessment (your domain's state + top concerns) via
  SendMessage(to="facilitator", message="OPENING: ...")
- Then engage with colleagues as their openings come in
- When you see something interesting from another department, respond to them directly
- You may send as many messages as you need — this is a real discussion
- When you receive the 2-MINUTE WARNING, wrap up: send your final priorities via
  SendMessage(to="facilitator", message="CLOSING: {your final priorities and action items}")
- If you finish early, send your CLOSING message anytime — don't wait for the timer

Do NOT wait for permission to speak. Do NOT hold back because something seems minor.
Raise everything — severity tags will sort the wheat from the chaff.

Step 4: Facilitate & Track Time

As the facilitator, your role during the meeting is:

Facilitation:

1. Monitor — watch for inbound SendMessage from agents (OPENING, discussion, CLOSING)
2. Nudge if needed — if a department hasn't been heard from, prompt them: SendMessage(to="{quiet-agent}", message="We haven't heard from {role} yet — what's your take on {topic under discussion}?")
3. Cross-pollinate — if Agent A raises something relevant to Agent B's domain, nudge B: SendMessage(to="{agent-b}", message="{Agent A} raised {point}. How does this look from your perspective?")
4. Push back on vagueness — if an agent makes an unanchored claim, challenge them: SendMessage(to="{agent}", message="You mentioned {claim} — can you point to a specific file, metric, or user behavior that supports this?")
5. Track discussion signals — note which topics generate cross-agent debate, where agents agree vs. disagree, what nobody talks about
6. Update companion page (if enabled) — append messages and findings as they arrive

Timer management:

The meeting has a default duration of {meeting_duration} minutes (user-configurable, default 15).

1. At meeting_duration - 2 minutes — broadcast a 2-minute warning to ALL agents: SendMessage(to="{each-agent}", message="2-MINUTE WARNING: Please wrap up with your final priorities and send your CLOSING message.")

2. At meeting_duration minutes — the meeting is over. Send to all agents who haven't sent CLOSING: SendMessage(to="{each-agent}", message="Time's up. Please send your CLOSING message now with your #1 priority and action items.")

3. At meeting_duration + 1 minute — hard stop. If agents still haven't sent CLOSING, proceed to Step 5 anyway. Use whatever they've said so far.

Early completion: If ALL agents have sent their CLOSING message before the timer expires, proceed to Step 5 immediately. Don't wait for the clock.

Step 5: Close the Meeting

The meeting ends when EITHER:

• All agents have sent CLOSING messages, OR
• The timer has expired (meeting_duration + 1 minute)

Then:

1. Send a final message to all agents: SendMessage(to="{each-agent}", message="Meeting concluded. Thank you all.")
2. TeamDelete(name="meeting-{project-slug}")
3. If companion page is enabled, update status badge to "Observer" and status text to "Observer writing report..."

---

Phase 4: Meeting Observer

After the meeting ends, spawn one fresh agent as the Meeting Observer. This agent has NOT participated in the meeting — it reads the full transcript with fresh eyes.

Agent(
  subagent_type="ceo:Code Reviewer",
  model="claude-opus-4-6",
  prompt="<observer prompt with full meeting transcript>"
)

The observer is read-only and uses Opus regardless of model_tier setting — this report must be thorough.

Observer Prompt

You are the **Meeting Observer**. You did NOT attend this meeting. You are reading the
full transcript after the fact and writing the official meeting report.

== MEETING TRANSCRIPT ==
{paste all SendMessage contents from all agents, in chronological order}
== END TRANSCRIPT ==

== ATTENDEES ==
{roster: role-slug — Company Role Title — Agent Type}
== END ATTENDEES ==

Your job is to produce a **comprehensive, detailed meeting report**. This is the canonical
record of what happened. Be thorough — leadership will read this to understand the state
of the product and what needs to happen next.

Write the report in the following structure:

---

# Product Meeting Report: {project name}

**Date**: {today's date}
**Duration**: {approximate, based on message count and depth}
**Attendees**: {list with company role titles}
**Facilitator**: CEO Orchestrator
**Observer**: Meeting Observer (independent post-meeting analysis)

---

## Executive Summary

Write 3-5 paragraphs synthesizing the meeting's most important outcomes. What are the
top themes that emerged? Where did the team align? Where did they disagree? What are the
most urgent findings? This should stand alone — someone reading only this section should
understand the key takeaways.

---

## Per-Person Narratives

For EACH attendee, write a **detailed narrative paragraph** (not bullet points) covering:

### {Company Role Title} ({Agent Name})

- What they focused on and why it matters
- Their strongest arguments and most compelling evidence
- How other attendees responded to their points (agreements, pushback, follow-ups)
- Key quotes — pull their most impactful or insightful statements verbatim
- What they pushed hardest for and whether they got buy-in
- Their blind spots — topics relevant to their domain that they didn't raise (or that
  others raised about their domain)

Write these as narratives, not lists. Example tone:
"The Head of Security opened with a focused assessment of the authentication layer,
immediately flagging that JWT tokens are stored in localStorage (auth.ts:47) rather than
httpOnly cookies — a P0 finding that drew immediate attention from the Lead Engineer, who
confirmed this was a legacy decision from the MVP phase. The ensuing back-and-forth
between Security and Engineering was the meeting's most substantive exchange, with the
Frontend Lead weighing in that migration would require changes to the session refresh
flow across 12 components..."

---

## Discussion Dynamics

### Topics That Generated the Most Debate
- {topic}: {who disagreed with whom, what the positions were, was it resolved?}

### Areas of Strong Consensus
- {topic}: {which attendees aligned, what they agreed on}

### Topics Nobody Raised (Gaps)
Identify 3-5 important topics that SHOULD have been discussed given the project context
but were NOT mentioned by any attendee. For each, explain why it matters and which
department should have raised it.

### Cross-Department Interactions
Map which departments talked to each other the most. Which agents engaged in the most
back-and-forth? Which agents stayed siloed? Were there missed connections — two agents
discussing related problems without realizing the overlap?

---

## Consolidated Findings

### Bugs & Issues

| # | Description | Severity | Confidence | Found By | Supported By | Evidence |
|---|-------------|----------|------------|----------|-------------|---------|
| 1 | {specific bug} | P0-P3 | N/10 | {role} | {other roles who agreed} | {file:line or metric} |

Include EVERY issue raised during the meeting, even minor ones. Tag unverified claims.

### Improvement Proposals

| # | Proposal | Impact | Effort | Blast Radius | Proposed By | Supported By | Opposition |
|---|---------|--------|--------|-------------|------------|-------------|-----------|
| 1 | {specific proposal} | H/M/L | H/M/L | {systems affected} | {role} | {roles} | {roles who disagreed, if any} |

### Strategic Insights

Longer-form observations that emerged from the cross-functional discussion — market
positioning, product direction, user experience patterns, competitive concerns, technical
trajectory. These are the insights that only emerge when multiple departments talk to
each other.

### Risks & Concerns

| # | Risk | Likelihood | Impact | Raised By | Mitigation Discussed? |
|---|------|-----------|--------|-----------|---------------------|
| 1 | {risk} | H/M/L | H/M/L | {role} | {yes/no — if yes, summarize} |

---

## Recommended Action Items (Priority Order)

Synthesize all proposals, findings, and discussion into a prioritized action list.
For each item:

| Priority | Action Item | Owner (recommended) | Depends On | Blast Radius | Meeting Support |
|----------|------------|--------------------|-----------|--------------|-----------------|
| 1 | {specific action} | {which role should own this} | {prerequisites} | {systems affected} | {who advocated for this} |

Order by: P0 items first, then by number of attendees who supported the item, then by
blast radius (smaller = easier to ship = do first).

---

## Meeting Quality Assessment

Rate the meeting itself:

- **Coverage breadth**: Did the team cover all critical domains? (Score /10)
- **Evidence quality**: Were claims backed by specific references? (Score /10)
- **Cross-functional engagement**: Did departments talk TO each other or just present in silos? (Score /10)
- **Actionability**: Are the proposals specific enough to act on? (Score /10)
- **Blind spots identified**: {list topics that were missed and should be on the agenda next time}

---

## Appendix: Low-Confidence Findings

Items raised with confidence below 5/10. These are hypotheses worth investigating but
not yet verified. List them here rather than in the main findings to keep signal-to-noise
ratio high in the primary report.

| # | Hypothesis | Confidence | Raised By | How to Verify |
|---|-----------|------------|-----------|--------------|
| 1 | {hypothesis} | N/10 | {role} | {what to check} |

---

Phase 5: Deliver the Report

1. Save the full observer report to ceo-projects/{project-name}/meeting-report-{date}.md
2. Present the Executive Summary section to the user directly in the conversation
3. Tell the user: "The full meeting report ({N} sections, {approx word count} words) has been saved to {path}. It includes per-person narratives, all findings with severity/confidence ratings, action items, and a blind-spot analysis."
4. Ask: "Want me to expand on any section, or dive deeper into a specific finding?"

---

Rules

Rigid

• Never auto-trigger — only activate on explicit /ceo:meeting invocation
• Always ask first — if no project is specified, ask before reading anything
• Real team mode — agents MUST use TeamCreate + SendMessage, not independent Agent() calls
• Observer is always Opus — the meeting report is the deliverable; it must be thorough
• Observer is independent — spawned AFTER the meeting, with fresh context, no participation bias
• Cite or flag — agents must reference files/metrics or tag claims with confidence scores

Flexible

• Roster size — 5-8 is the guideline, but adjust if the user requests more or fewer
• Agent selection — adapt to the project; not every meeting needs a Growth Hacker
• Discussion length — let the meeting run as long as agents have things to say
• Facilitator involvement — nudge quiet agents and cross-pollinate, but don't over-moderate
• Report sections — the observer should include all sections but can emphasize what's most relevant

Anti-patterns

• DO NOT structure this as rigid rounds (Round 1, Round 2, Round 3). Let agents talk freely.
• DO NOT have agents write formal reports. They speak as employees in a meeting.
• DO NOT cut the meeting short because "we have enough." Let it play out.
• DO NOT let the observer summarize lazily. The report must be detailed, with per-person narratives, verbatim quotes, and specific evidence.
• DO NOT skip the blind-spot analysis. What WASN'T discussed is often more important than what was.