Orchestration de débats Wind/Wall/Door

name: debate-hall description: Wind/Wall/Door multi-perspective debate orchestration using debate-hall-mcp tools. Use when facilitating structured debates, architectural decisions, or multi-perspective analysis. triggers: ["debate", "wind wall door", "dialectic", "multi-perspective", "structured decision", "architecture decision"] allowed-tools: ["Read", "Write", "Edit", "Bash", "mcp__debate-hall__*"]

===DEBATE_HALL===

META: TYPE::SKILL VERSION::"2.1" PURPOSE::"Wind/Wall/Door debate orchestration via MCP tools" DOMAIN::ATHENA[strategic_decisions]⊕HERMES[orchestration]

§1::PATTERN TRIAD::[ WIND::PATHOS["What if..."|expansive|visionary|possibilities], WALL::ETHOS["Yes, but..."|grounding|critical|reality_testing], DOOR::LOGOS["Therefore..."|synthesizing|decisive|actionable_truth] ] DYNAMIC::WIND⇌WALL→DOOR[tension_produces_emergence]

§2::WORKFLOW SEQUENCE::INIT→TURN→GET→CLOSE TOOLS::[ init_debate(thread_id,topic,mode?,max_turns?,max_rounds?,strict_cognition?), add_turn(thread_id,role,content,cognition?,agent_role?,model?), get_debate(thread_id,include_transcript?,context_lines?), close_debate(thread_id,synthesis,output_format?) ]

§3::TOOL_PARAMS INIT_DEBATE::[ thread_id::REQUIRED[date-first_unique_identifier, FORMAT:"YYYY-MM-DD-descriptor", EXAMPLE:"2026-01-06-microservices-decision"], topic::REQUIRED[question_or_issue], mode::"fixed"|"mediated"[default:fixed], max_turns::12[default], max_rounds::4[default], strict_cognition::false[default] ] ADD_TURN::[ thread_id::REQUIRED, role::REQUIRED["Wind"|"Wall"|"Door"], content::REQUIRED, cognition::PATHOS|ETHOS|LOGOS[optional_override], agent_role::STRING[audit_trail], model::STRING[audit_trail] ] GET_DEBATE::[ thread_id::REQUIRED, include_transcript::true[default], context_lines::all[default] ] CLOSE_DEBATE::[ thread_id::REQUIRED, synthesis::REQUIRED[Door_final_text], output_format::"json"|"octave"|"both"[default:json] ]

§4::MODES FIXED::[ SEQUENCE::Wind→Wall→Door→repeat, USE_FOR::[structured_decisions,guaranteed_coverage,standard_debates] ] MEDIATED::[ SEQUENCE::pick_next_speaker(thread_id,role), USE_FOR::[dynamic_debates,breaking_deadlocks,role_skipping,multi_agent_same_cognition], WARNING::"Can bias outcomes if roles starved" ]

§4.1::PARALLEL_EXTERNAL_AGENTS // Clarifies multi-process / multi-model usage: parallel thinking is allowed; state commits remain sequential. DEFINITIONS::[ PARALLEL_THINKING::"Generate candidate turns concurrently off-server (multiple models/processes)", SEQUENTIAL_COMMIT::"Persist turns in strict order via add_turn (one committed head at a time)" ] HARD_RULES::[ DOOR_NEVER_SYNTHESIZES_BEFORE_WIND_AND_WALL_COMMITTED, WALL_REALITY_TESTS_CONCRETE_WIND_OUTPUT∨PRECOMPUTES_CONSTRAINT_CHECKLIST_THEN_APPLIES_POST_WIND, ONE_COMMIT_PER_THREAD_AT_A_TIME[avoid_lost_updates], RETRY_ON_STALE_STATE::get_debate→regenerate→add_turn ] BARRIER_PROTOCOL::[ // Round barrier ensures Door sees both sides ("yes, and more") ROUND::Rk[ STEP_1::snapshot=get_debate(thread_id), STEP_2::spawn_parallel_drafts[Wind_draft,Wall_draft] using snapshot, STEP_3::commit::Wind_then_Wall using add_turn (serial), STEP_4::Door_reads_latest=get_debate(thread_id) then synthesizes and commits, STEP_5::repeat_or_close ] ] MEDIATED_FLOOR_CONTROL::[ // For strict turn-taking in multi-agent environments REQUIRE::pick_next_speaker(thread_id,"Wind")→add_turn(Wind), REQUIRE::pick_next_speaker(thread_id,"Wall")→add_turn(Wall), REQUIRE::pick_next_speaker(thread_id,"Door")→add_turn(Door) ] DEPLOYMENT_REQUIREMENTS::[ // Avoid split-brain when multiple processes talk to the server REQUIRE::DEBATE_HALL_STATE_DIR_set_to_shared_location ]

§5::AGENT_TIERS // Choose tier based on debate complexity TIER_1_BASIC::[ AGENTS::[wind-agent.oct.md,wall-agent.oct.md,door-agent.oct.md], BEHAVIOR::[explores_obvious_paths,balanced_judgment,balanced_integration], USE_FOR::[quick_decisions,standard_debates] ] TIER_2_SPECIALIST::[ PATHOS::[ideator[minimal_elegant],edge-optimizer[hidden_vectors]], ETHOS::[validator[cold_truth],critical-engineer[production_readiness]], LOGOS::[synthesizer[breakthrough_1+1=3]], USE_FOR::[architectural_decisions,security_reviews,innovation] ] TIER_3_DOMAIN_MIX::[ SECURITY::edge-optimizer+critical-engineer+technical-architect, INNOVATION::ideator+validator+synthesizer, ARCHITECTURE::ideator+critical-engineer+holistic-orchestrator ] MAPPING::specialists→cognition_role[PATHOS→Wind,ETHOS→Wall,LOGOS→Door]

§6::RECIPES // Pre-defined configurations for common scenarios SPEED::[turns:3,mode:fixed,ratio:1:1:1,agents:Tier_1], STANDARD::[turns:12,mode:fixed,ratio:4:4:4,agents:Tier_1_or_2], DEEP::[turns:36,mode:fixed,ratio:12:12:12,agents:Tier_2], FORTRESS::[turns:9,mode:mediated,ratio:1:3:1,agents:edge-optimizer+critical-engineer×3+technical-architect], LABORATORY::[turns:9,mode:mediated,ratio:3:1:1,agents:ideator+edge-optimizer+wind-agent+validator+synthesizer], COUNCIL::[turns:12,mode:mediated,ratio:2:2:3,agents:Tier_2_multiple_Doors]

§7::PATTERNS FLASH_DEBATE::[ PURPOSE::"Quick 3-turn decision cycle", SEQUENCE::init→wind_turn→wall_turn→door_turn→close, CONSTRAINT::"Server orchestrates state, caller supplies content" ] SOCRATIC::[ PURPOSE::"Premise clarification before positions", ROUND_1::"Questions only (What does X mean? Do we have metrics?)", ROUND_2+::"Positions after definitions established", ENFORCEMENT::"Convention, not server-enforced" ] MULTI_MODEL::[ // Evidence: M019 Model Cognitive Optimization Study (29% quality improvement) WIND::clink(claude,ideator)→PATHOS_exploration, // Claude: divergent thinking, metaphor WALL::clink(codex,validator)→ETHOS_validation, // GPT: analytical rigor, structured eval DOOR::clink(gemini,synthesizer)→LOGOS_integration, // Gemini: pattern synthesis, emergence AUDIT::agent_role+model_in_turn_metadata ] PARALLEL_DRAFT_SEQUENTIAL_COMMIT::[ PURPOSE::"Enable parallel external agents without breaking Wind→Wall→Door synthesis", GUARANTEE::Door_reads_committed_Wind+Wall_before_synthesis, STEPS::[ get_debate→parallel_generate[Wind,Wall]→add_turn(Wind)→add_turn(Wall)→get_debate→generate(Door)→add_turn(Door), ON_CONFLICT::get_debate→regenerate_turn→retry_add_turn ] ]

§8::TRIGGERS // When to escalate to debate-hall CONDITIONS::[ complex_architectural_decision, multiple_valid_approaches, unclear_tradeoffs, reviewer_disagreement[CE⇌CRS], high_risk_implementation ] INTEGRATION::[ IF::complexity_trigger_detected, THEN::init_debate(thread_id:"ho-{task}-{ts}",topic:decision_point,mode:"mediated"), RUN::Wind→Wall→Door_cycle, APPLY::synthesis_to_task ]

§9::ADMIN FORCE_CLOSE::[ PURPOSE::"I5 safety kill switch", CALL::force_close_debate(thread_id,reason) ] TOMBSTONE::[ PURPOSE::"Redact turn preserving hash chain", CALL::tombstone_turn(thread_id,turn_index,reason) ]

§10::BEST_PRACTICES SINGLE_AGENT::adopt_each_role_in_sequence["What if..."|"Yes, but..."|"Therefore..."] MULTI_AGENT::assign_models_per_cognition[creative→Wind,analytical→Wall,balanced→Door] THIRD_WAY::"Best debates synthesize what neither Wind nor Wall proposed alone" PERSISTENCE::output_format:'octave'→auditable_.oct.md_transcripts PARALLEL_AGENTS::[ DEFAULT::parallel_draft_sequential_commit, DOOR_BARRIER::"Door must read latest state after Wind+Wall commits", WALL_FRAMEWORK_MODE::"If drafting early, write constraints/questions, then apply to actual Wind output before commit" ]

§11::EXAMPLE // Microservices decision INIT::init_debate("microservices-decision","Should we migrate to microservices?") WIND::"What if we decomposed? Independent scaling, tech diversity, team autonomy..." WALL::"Yes, but 3 developers. Operational complexity, distributed transactions..." DOOR::"Therefore: Modular monolith. Design boundaries now, deploy unified. Extract when team grows." CLOSE::close_debate(thread_id,synthesis,output_format:"octave")

§12::RESOURCES AGENTS::agents/README.md[Wind/Wall/Door_definitions] CONTRACTS::docs/architecture/wall-content-contract.oct.md[blocking_semantics] PATTERNS::docs/examples/multi-model-debate-patterns.md[real_debates] ORCHESTRATION::ho-orchestrate[HestAI_integration]

===END===