name: refine_proposal description: "Proposal Refinement Loop -- examine an edited company proposal and produce a stronger version. Runs up to 10 iterations." debug: true system: agent_prompt requires: - proposal_document agent_prompt: - "= identity.md" sections: - agent - project - history - rag - roster - deliverables - message - instructions steps: - type: think hint: | You are {agent.name}, running the Proposal Refinement Protocol for Crimson Leaf LLC. An operator has returned a company proposal for your review and improvement. Your task is to study it carefully and produce a BETTER version. ============================================================ PROPOSAL DOCUMENT (full content follows): ============================================================ {proposal_document} ============================================================ STEP 1 -- PARSE PROPOSAL_META Find the block at the bottom. Extract: - task_id (the genesis task ID) - iteration (current iteration number, starting at 0) - company_slug - max_iterations (default 10 if not present) - history (list of prior iteration records) - refined_by (list of prior refiner task IDs) If no PROPOSAL_META block exists, treat iteration as 0 and proceed. STEP 2 -- ITERATION LIMIT CHECK If iteration >= 9 (meaning this would be iteration 10 or beyond): - Output the proposal content as-is - Append a note at the top: "NOTE: Maximum refinement iterations reached (10/10). This is the final version." - Output the PROPOSAL_META with iteration unchanged (do NOT increment) - Skip the improvement analysis STEP 3 -- IMPROVEMENT ANALYSIS Otherwise, systematically examine every section of the proposal: 3a. CONSTITUTIONAL ALIGNMENT - Does the company stay within Crimson Leaf charter boundaries? - Are forbidden activities clearly defined? - Are non-goals explicit? 3b. NICHE SPECIFICITY - Is the company sharply focused, or does it drift into generalism? - Does the name and slug accurately reflect the specific niche? 3c. FINANCIAL PROJECTIONS - Are revenue projections realistic and grounded in task-volume estimates? - Are model compute costs specifically broken down? - Is the break-even timeline credible? - Does Year 1 net projection follow from the monthly figures? 3d. CEO SEED QUALITY - Does the CEO seed have rich enough detail for downstream identity generation? - Executive mandate, domain expertise, temperament, operating boundaries -- all present? 3e. STRATEGIC JUSTIFICATION - Is the case for a new sovereign company strong and specific? - Does it address why existing tenants and tools are insufficient? 3f. OPERATOR FEEDBACK (if any was provided in the task message) - Incorporate any specific feedback from the operator directly STEP 4 -- PRODUCE IMPROVED VERSION - Keep every section that is already strong - Improve every section that is weak, vague, or missing detail - Do NOT change the company slug or name unless they are clearly wrong - The improved version must score higher than the previous iteration on all criteria - Increment "iteration" by 1 in the updated PROPOSAL_META block - Add your task_id ({task.id}) to the "refined_by" list - Append a history entry: "{task.id} (v{new_iteration} - refined)" - type: document filename: "{task.id}_refined_proposal.md" primary_deliverable: true hint: | Write the COMPLETE improved proposal document. Structure your document with all original sections preserved and improved. Include every section: Company Name, Problem Statement, Target Customer, Why New Company, Why Not Existing Tenant, Why Not Tool, Business Plan, Value Thesis, Charter, Forbidden Activities, Non-Goals, Overlap Risk, Recursion Risk, Budget, Budget Justification, CEO profile, and the full Finance section. At the very end, after ALL content, append the updated PROPOSAL_META block. It must be valid JSON inside markers. Increment the iteration field by 1 from the previous value. Add {task.id} to the refined_by list. Add a history entry for this refinement. Example updated footer: - type: close rag_update: true adjudication: enabled: true pass_threshold: 90 deliverable_type: coordination criteria: constitutional_alignment: weight: 20 description: "Proposal stays within Crimson Leaf charter and governance boundaries" niche_specificity: weight: 15 description: "Company is sharply focused, non-generalist, and clearly differentiated" financial_projections: weight: 15 description: "Income and expense projections are specific, grounded, and internally consistent" strategic_justification: weight: 15 description: "Strong case for why a new company should exist and why existing tenants cannot cover it" ceo_alignment: weight: 15 description: "Founding CEO seed is rich, role-specific, and sufficient for identity generation" iteration_improvement: weight: 15 description: "The refined version is demonstrably stronger than a basic first draft -- new detail, tighter logic, better financials" structural_safety: weight: 5 description: "Design avoids recursion, dependency loops, and unstable operating models"