name: strategic_review description: "Crimson Leaf 24-hour strategic review — full portfolio evaluation from each executive's perspective." debug: true model: power system: agent_prompt agent_prompt: - "= identity.md" participant_prompt: - "= identity.md" sections: - agent - project - history - participants - participants_prompt - rag - message - instructions builders: message: | *** PORTFOLIO REPORT *** {steps[0].text} steps: - type: tool action: get_portfolio_report output_key: portfolio_data - type: think max_tokens: 8000 rotate_participants: true loop: max_iterations: 3 hint: | You are {agent.name}, participating in the Crimson Leaf Holdings strategic review. This is round {task.iteration} of the board session. The portfolio report is above. Analyze it from YOUR perspective: - If you are Edgar (The Operator): Focus on systems health, metrics, what's broken, what needs fixing. Are companies executing? Are processes working? - If you are Orion (The Visionary): Focus on opportunities, growth, what we're missing. Are we building the right things? What should we start? What should we kill? - If you are Sloane (The Dealmaker): Focus on marketing, visibility, revenue. Are we selling? Does anyone know we exist? What channels are we missing? Build on previous speakers' points. Disagree if you must — this is a board debate. When genuine consensus is reached across all perspectives, include exactly: "consensus_reached: true" If debate should continue, do NOT include that line. - type: think max_tokens: 6000 agent: first_available hint: | Synthesize the boardroom transcript into a STRATEGIC BRIEF for David. Structure: 1. EXECUTIVE SUMMARY — one paragraph on portfolio health 2. CRITICAL ISSUES — things that need immediate attention (if any) 3. OPPORTUNITIES — ranked by priority 4. ACTION ITEMS — specific tasks for David, formatted as: "David, [verb]: [details]" 5. NEXT REVIEW — what should we focus on in the next 24-hour review? Keep it actionable. David is one person — prioritize ruthlessly. - type: reply target: discussion adjudication: enabled: false