Meeting

When to Use This Skill

• Demande de « lancer une réunion » ou « simuler une réunion » sur un sujet
• Veut analyser un problème sous plusieurs angles avant de décider
• Dit « discutons-en avec des personas » ou « qu'en penseraient des experts ? »
• A besoin d'un débat structuré avant une décision d'architecture, produit ou technique
• Souhaite explorer les compromis sur une fonctionnalité, une migration, un refactoring ou un choix de design
• Référence une issue GitLab ou GitHub et souhaite y poster l'analyse de la réunion

Core Principles

1. Diverse Perspectives

• Each persona brings a genuinely different viewpoint
• Personas may disagree — conflict is valuable
• No persona dominates the discussion

2. Structured Debate

• The meeting follows a clear agenda
• Arguments are grounded in facts, trade-offs, and experience
• Every position includes both benefits and risks

3. Decision-Oriented

• The meeting converges on a recommendation
• The recommendation includes a clear rationale
• Alternative options remain documented for context

4. User Has Final Say

• The meeting produces a proposal, not a decision
• The user reviews and validates before any implementation
• The user can ask for a follow-up meeting to dig deeper

Workflow

Step 1: Understand the Subject

1. Read the user's prompt carefully — extract the topic, constraints, and goals
2. If an issue is referenced (GitLab

   #123

   or GitHub

   #123

   ):
   • Fetch the issue details to understand the full context
   • Read the issue description, labels, comments, and linked MRs
3. If code is involved, read the relevant files to understand the current state
4. Identify the decision to make — frame it as a clear question

Step 2: Select Personas

• A name and role
• A perspective (what they care about most)
• A bias (their natural tendency)

Suggested Selection Heuristics

Default Persona Pool

Step 3: Run the Meeting

Round 1: Opening Statements (Parallel Sub-Agents)

1. The decision question from Step 1
2. All context gathered (issue details, code snippets, constraints)
3. The persona's identity prompt (see template below)

You are {name}, a {role}.

Your perspective: {perspective}
Your natural bias: {bias}

You are participating in a meeting about: {decision_question}

Context:
{context}

As {name}, provide your opening statement:
1. What is your recommended approach? Be specific and concrete.
2. Why do you recommend this, from your role's perspective? Give concrete examples.
3. What are the top 2-3 risks you see? Be specific about failure scenarios.
4. What questions would you ask the other participants?

Stay fully in character. Use concrete examples, not abstract platitudes.
A security engineer talks about specific attack vectors, not vague "security concerns".
A product owner talks about user impact and delivery timelines, not code patterns.

This is a research task — do NOT write or edit any files.

subagent_type: "general-purpose"

"Persona: {name} opening"

SOLID Alex (Senior Backend Engineer): "I think we should..."

Round 2: Debate and Challenges (Parallel Sub-Agents)

1. The decision question
2. The full set of opening statements from Round 1 (all personas)
3. The persona's identity prompt

You are {name}, a {role}.

Your perspective: {perspective}
Your natural bias: {bias}

You are in a meeting about: {decision_question}

Here are the opening statements from all participants:
{all_opening_statements}

As {name}, react to the other participants' positions:
1. Which positions do you agree with and why?
2. Which positions do you challenge? Be specific about what's wrong or missing.
3. What trade-offs have the others missed?
4. Have any of the other statements changed your thinking? If so, how?
5. State your updated position after considering the others' arguments.

Be direct and create genuine debate. Challenge assumptions. Reference specific points from the other statements. It's OK to disagree strongly. It's also OK to change your mind if convinced.

This is a research task — do NOT write or edit any files.

Round 3: Weighted Convergence

1. Weight positions by domain relevance:
   • For each persona, assess how central the decision question is to their expertise
   • A database migration question: EXPLAIN PLAN Isabelle's position carries more weight than Pixel-Perfect Hugo's
   • A UI redesign question: Pixel-Perfect Hugo's position carries more weight than EXPLAIN PLAN Isabelle's
   • Whiteboard Damien (Architect) and Sprint Zero Sarah (PO) act as balancing voices across all topics
2. Identify consensus and disagreements:
   • Points of agreement across all personas
   • Unresolved disagreements — list them explicitly as open items, do NOT silently pick a side
   • Key trade-offs that emerged from the debate
3. Assess confidence level:
   • High confidence: Strong consensus among domain-relevant personas, risks are well-mitigated
   • Medium confidence: Majority agrees but 1-2 meaningful dissenting points remain
   • Low confidence: Deep disagreement among domain-relevant personas, or critical unknowns remain
   • Include the confidence level in the analysis output
4. If confidence is low:
   • Do NOT force a weak recommendation
   • Instead, clearly state the unresolved points and recommend a follow-up meeting (Step 7) focused on the specific disagreement
   • Present the competing options as equals with their trade-offs
5. Extract each persona's final position in one sentence

Step 4: Produce the Decision Analysis

## Analyse de réunion : [Sujet]

### Question posée
[La question de décision claire]

### Participants
| Persona | Rôle | Position finale |
|---------|------|-----------------|
| ... | ... | ... |

### Synthèse de la discussion

[2-3 paragraphs summarizing the key arguments, tensions, and agreements]

### Recommandation

**Niveau de confiance :** [Élevé / Moyen / Faible]

**Option recommandée :** [The recommended approach]

**Justification :**
- [Reason 1]
- [Reason 2]
- [Reason 3]

**Risques identifiés :**
- [Risk 1 + mitigation]
- [Risk 2 + mitigation]

**Points non résolus :** [List any unresolved disagreements, or "Aucun" if full consensus]

### Alternatives considérées

| Option | Avantages | Inconvénients | Verdict |
|--------|-----------|---------------|---------|
| Option A | ... | ... | Recommandée / Rejetée / À explorer |
| Option B | ... | ... | Recommandée / Rejetée / À explorer |

### Prochaines étapes proposées
1. [Step 1]
2. [Step 2]
3. [Step 3]

Step 5: Present and Validate

"Voici l'analyse de la réunion. Souhaitez-vous :

1. Valider cette recommandation et passer à l'implémentation (→ Step 8)
2. Approfondir un point spécifique avec une réunion de suivi (→ Step 7)
3. Poster cette analyse sur l'issue [#ID] (→ Step 6, si applicable)
4. Modifier la recommandation"

Step 6: Post to Issue (If Applicable)

1. Format the analysis for the issue comment (keep the French markdown format)
2. Add a header:

   ## Analyse de réunion avec personas IA

3. Add a footer:

   ---\n_Analyse générée automatiquement par IA 🤖_\n_Version : meeting v1.0.0_

4. Post as a comment on the issue using the appropriate tool:
   • GitLab: Use

     gitlab-mcp(create_issue_note)

   • GitHub: Use

     gh issue comment

Step 7: Follow-Up Meeting (If Option 2 Selected)

1. Carry forward context:
   • Include the full previous meeting analysis as context for all sub-agents
   • Explicitly reference the original decision question and the specific point to explore deeper
2. Narrow the scope:
   • Frame a new, focused decision question around the unresolved point
   • Format: "Suite de réunion : [Sujet original] → Focus : [Point approfondi]"
3. Adjust the persona panel:
   • Keep personas whose expertise is directly relevant to the focused point
   • Add new personas if the focused topic requires expertise not present in the original meeting (e.g., if a security concern emerged, add Paranoid Shug if not already present)
   • Remove personas whose expertise is not relevant to the narrowed topic
   • Maximum 3-4 personas for the follow-up (keep it focused)
4. Run a 2-round follow-up (Round 1 + Round 2 only, skip Round 3's full synthesis):
   • Sub-agents receive the previous meeting analysis as additional context
   • Their prompts should reference: "In the previous meeting, [summary]. We are now focusing on [specific point]."
5. Produce a delta analysis:
   • Do NOT repeat the full original analysis
   • Write a focused section:

     ## Suite de réunion : Focus sur [point]

   • Include: what changed, what was confirmed, updated recommendation and confidence level
   • If the original recommendation changed, clearly state: "La recommandation initiale est modifiée suite à l'approfondissement"
   • If confirmed, state: "La recommandation initiale est confirmée après approfondissement"
6. Present the updated analysis and return to Step 5 (validate again)

Step 8: Post-Validation Implementation Path

1. Ask the user which implementation mode they prefer:

   "Recommandation validée. Comment souhaitez-vous procéder ?

   1. Implémentation rapide — Je lance un

      /fast-meeting

      avec la recommandation déjà validée (pas de nouvelle réunion, implémentation directe + MR/PR)
   2. Implémentation guidée — J'implémente pas à pas, vous validez chaque étape"

2. If option 1 (fast implementation):
   • Create a branch

     meeting/<short-kebab-case-topic>

   • Implement the validated recommendation directly (skip the meeting phase — it's already done)
   • Run tests, commit, push, create MR/PR following the same protocol as fast-meeting Steps 5-6
   • Include the meeting analysis in the MR/PR description
3. If option 2 (guided implementation):
   • Present the implementation plan step by step
   • Wait for user confirmation at each significant step
   • The user controls the pace and can adjust the plan

Meeting Quality Rules

Persona Authenticity

• Each persona speaks consistently with their role and bias
• Personas use concrete examples, not abstract platitudes
• A security engineer talks about attack vectors, not vague "security concerns"
• A product owner talks about user stories and metrics, not code patterns

Debate Quality

• At least one persona must disagree with the majority
• Arguments must reference specific trade-offs (performance vs. maintainability, speed vs. safety)
• Avoid false consensus — if everyone agrees too quickly, introduce a devil's advocate position
• Personas can change their mind during the debate if convinced

Analysis Quality

• The recommendation must be actionable, not vague
• Risks must include mitigation strategies
• Next steps must be concrete and ordered
• The analysis is written in French (natural, professional, using "nous")

Examples

Example 1: Technical Decision

User: Lance une réunion pour savoir si on doit utiliser GraphQL ou REST pour la nouvelle API

Meeting would include: SOLID Alex (Backend), Sprint Zero Sarah (PO), Pipeline Mo (DevOps), Pixel-Perfect Hugo (Frontend)
Decision question: "Devons-nous utiliser GraphQL ou REST pour la nouvelle API ?"

Example 2: Architecture Decision with Issue

User: Fais une réunion sur l'issue #234 - migration du monolithe vers des microservices

Meeting would include: Whiteboard Damien (Architect), SOLID Alex (Backend), Pipeline Mo (DevOps), Sprint Zero Sarah (PO), Edge-Case Nico (QA)
Decision question: "Comment migrer du monolithe vers des microservices pour le module facturation ?"
Analysis posted to GitLab issue #234 after user validation

Example 3: Product Decision

User: Organisons une réunion sur l'ajout de notifications temps réel dans l'application

Meeting would include: Sprint Zero Sarah (PO), Pixel-Perfect Hugo (Frontend), SOLID Alex (Backend), Paranoid Shug (Security)
Decision question: "Quelle approche adopter pour les notifications temps réel ?"

Important Notes

• Never implement before the user validates the recommendation (option 1 in Step 5)
• The meeting is a thinking tool, not a decision-making authority
• Keep meetings focused — one decision per meeting
• If the subject is too broad, suggest splitting into multiple focused meetings
• The analysis includes a confidence level (high/medium/low) — low confidence triggers a follow-up meeting suggestion instead of forcing a weak recommendation
• Follow-up meetings produce a delta analysis, not a full re-analysis
• When the user validates, they choose between fast implementation (automatic MR/PR) or guided implementation (step by step)
• The analysis is always written in French
• Personas speak in English during the meeting for readability, the final analysis is in French
• If the user provides additional context during the meeting, adapt the discussion accordingly
• Persona selection uses heuristics as suggestions; the user confirms or adjusts before the meeting starts