claude-flow-novice 2.6.0 → 2.8.1

package/.claude/skills/redis-coordination/cfn-loop-exec.sh

@@ -0,0 +1,468 @@
+#!/usr/bin/env bash
+##############################################################################
+# CFN Loop Executor - Self-Contained Orchestration
+#
+# Purpose: Eliminates need for coordinator agent by making orchestration
+#          fully deterministic and self-contained in CLI layer.
+#
+# Usage:
+#   ./cfn-loop-exec.sh --task "Build React dashboard" [options]
+#
+# What this does:
+#   1. Analyzes task description
+#   2. Selects optimal agents using registry
+#   3. Spawns orchestrator in background
+#   4. Monitors via Redis
+#   5. Returns structured JSON result
+#
+# Benefits:
+#   - No coordinator agent needed (more cost savings)
+#   - Fully deterministic (no LLM interpretation)
+#   - Direct Main Chat → orchestrator
+#   - JSON output for easy parsing
+##############################################################################
+
+set -euo pipefail
+
+# Configuration
+TASK_DESCRIPTION=""
+MODE="standard"
+OUTPUT_FORMAT="json"
+MAX_ITERATIONS=10
+VERBOSE=false
+WAIT_FOR_COMPLETION=true
+DIFFICULTY="auto"  # auto | simple | standard | complex | enterprise
+
+# Parse arguments
+while [[ $# -gt 0 ]]; do
+  case $1 in
+    --task)
+      TASK_DESCRIPTION="$2"
+      shift 2
+      ;;
+    --mode)
+      MODE="$2"
+      shift 2
+      ;;
+    --difficulty)
+      DIFFICULTY="$2"
+      shift 2
+      ;;
+    --output)
+      OUTPUT_FORMAT="$2"
+      shift 2
+      ;;
+    --max-iterations)
+      MAX_ITERATIONS="$2"
+      shift 2
+      ;;
+    --verbose)
+      VERBOSE=true
+      shift
+      ;;
+    --background)
+      WAIT_FOR_COMPLETION=false
+      shift
+      ;;
+    *)
+      echo "Unknown option: $1"
+      exit 1
+      ;;
+  esac
+done
+
+# Validation
+if [ -z "$TASK_DESCRIPTION" ]; then
+  echo "Error: --task required"
+  echo "Usage: $0 --task \"Build React dashboard\" [--mode standard] [--output json]"
+  exit 1
+fi
+
+##############################################################################
+# Step 1: Analyze Task Complexity
+##############################################################################
+
+log() {
+  if [ "$VERBOSE" = true ]; then
+    echo "[CFN-EXEC] $*" >&2
+  fi
+}
+
+log "Analyzing task complexity: $TASK_DESCRIPTION"
+
+# Get script directory
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+
+# Run complexity analyzer
+COMPLEXITY_RESULT=$("$SCRIPT_DIR/analyze-task-complexity.sh" \
+  --task "$TASK_DESCRIPTION" \
+  --difficulty "$DIFFICULTY")
+
+# Extract complexity metrics
+CALCULATED_DIFFICULTY=$(echo "$COMPLEXITY_RESULT" | jq -r '.difficulty')
+COMPLEXITY_SCORE=$(echo "$COMPLEXITY_RESULT" | jq -r '.complexity_score')
+SUGGESTED_LOOP3=$(echo "$COMPLEXITY_RESULT" | jq -r '.suggested_agents.loop3_count')
+SUGGESTED_LOOP2=$(echo "$COMPLEXITY_RESULT" | jq -r '.suggested_agents.loop2_count')
+DETECTED_DOMAINS=$(echo "$COMPLEXITY_RESULT" | jq -r '.domains[]' | paste -sd ',' -)
+
+log "Complexity: $CALCULATED_DIFFICULTY (score: $COMPLEXITY_SCORE)"
+log "Suggested agents - Loop 3: $SUGGESTED_LOOP3, Loop 2: $SUGGESTED_LOOP2"
+log "Detected domains: $DETECTED_DOMAINS"
+
+##############################################################################
+# Step 2: Select Agents Based on Domain & Complexity
+##############################################################################
+
+# Convert task to lowercase for matching
+TASK_LOWER=$(echo "$TASK_DESCRIPTION" | tr '[:upper:]' '[:lower:]')
+
+# Initialize agent lists with base validators
+LOOP3_AGENTS=""
+LOOP2_AGENTS="reviewer,tester"
+
+# Track how many agents we need per domain
+LOOP3_COUNT=0
+LOOP3_MAX=$SUGGESTED_LOOP3
+
+# Frontend detection (can add multiple frontend specialists if complex)
+if [[ "$TASK_LOWER" =~ react|component|ui|frontend|dashboard|web.*app ]]; then
+  LOOP3_AGENTS="react-frontend-engineer"
+  LOOP3_COUNT=$((LOOP3_COUNT + 1))
+  LOOP2_AGENTS="${LOOP2_AGENTS},accessibility-advocate"
+  log "Added: Frontend specialist"
+
+  # Add UI designer for complex frontend work
+  if [ "$CALCULATED_DIFFICULTY" = "complex" ] || [ "$CALCULATED_DIFFICULTY" = "enterprise" ]; then
+    if [ $LOOP3_COUNT -lt $LOOP3_MAX ]; then
+      LOOP3_AGENTS="${LOOP3_AGENTS},ui-designer"
+      LOOP3_COUNT=$((LOOP3_COUNT + 1))
+      log "Added: UI designer (complex frontend)"
+    fi
+  fi
+fi
+
+# Backend detection
+if [[ "$TASK_LOWER" =~ api|backend|server|endpoint|rest|graphql|database|auth ]]; then
+  if [ $LOOP3_COUNT -lt $LOOP3_MAX ]; then
+    if [ -n "$LOOP3_AGENTS" ]; then
+      LOOP3_AGENTS="${LOOP3_AGENTS},backend-dev"
+    else
+      LOOP3_AGENTS="backend-dev"
+    fi
+    LOOP3_COUNT=$((LOOP3_COUNT + 1))
+    LOOP2_AGENTS="${LOOP2_AGENTS},security-specialist"
+    log "Added: Backend specialist"
+  fi
+fi
+
+# Database specialist for complex data work
+if [[ "$TASK_LOWER" =~ database|schema|migration|query|sql ]]; then
+  if [ "$CALCULATED_DIFFICULTY" = "complex" ] || [ "$CALCULATED_DIFFICULTY" = "enterprise" ]; then
+    if [ $LOOP3_COUNT -lt $LOOP3_MAX ]; then
+      LOOP3_AGENTS="${LOOP3_AGENTS},backend-dev"  # Additional backend specialist for DB work
+      LOOP3_COUNT=$((LOOP3_COUNT + 1))
+      log "Added: Database specialist (complex data work)"
+    fi
+  fi
+fi
+
+# Rust detection
+if [[ "$TASK_LOWER" =~ rust|cargo|tokio ]]; then
+  if [ $LOOP3_COUNT -lt $LOOP3_MAX ]; then
+    if [ -n "$LOOP3_AGENTS" ]; then
+      LOOP3_AGENTS="${LOOP3_AGENTS},rust-developer"
+    else
+      LOOP3_AGENTS="rust-developer"
+    fi
+    LOOP3_COUNT=$((LOOP3_COUNT + 1))
+    log "Added: Rust specialist"
+  fi
+
+  # Add enterprise Rust developer for production work
+  if [ "$CALCULATED_DIFFICULTY" = "enterprise" ]; then
+    if [ $LOOP3_COUNT -lt $LOOP3_MAX ]; then
+      LOOP3_AGENTS="${LOOP3_AGENTS},rust-enterprise-developer"
+      LOOP3_COUNT=$((LOOP3_COUNT + 1))
+      log "Added: Enterprise Rust specialist"
+    fi
+  fi
+fi
+
+# Infrastructure detection
+if [[ "$TASK_LOWER" =~ infra|devops|deploy|docker|k8s|kubernetes|aws ]]; then
+  if [ $LOOP3_COUNT -lt $LOOP3_MAX ]; then
+    if [ -n "$LOOP3_AGENTS" ]; then
+      LOOP3_AGENTS="${LOOP3_AGENTS},devops-engineer"
+    else
+      LOOP3_AGENTS="devops-engineer"
+    fi
+    LOOP3_COUNT=$((LOOP3_COUNT + 1))
+    log "Added: DevOps specialist"
+  fi
+fi
+
+# Architecture detection (always add for complex tasks)
+if [[ "$TASK_LOWER" =~ architect|design|system|scalab|pattern ]] || \
+   [ "$CALCULATED_DIFFICULTY" = "complex" ] || [ "$CALCULATED_DIFFICULTY" = "enterprise" ]; then
+  if [ $LOOP3_COUNT -lt $LOOP3_MAX ]; then
+    if [ -n "$LOOP3_AGENTS" ]; then
+      LOOP3_AGENTS="${LOOP3_AGENTS},system-architect"
+    else
+      LOOP3_AGENTS="system-architect"
+    fi
+    LOOP3_COUNT=$((LOOP3_COUNT + 1))
+    LOOP2_AGENTS="${LOOP2_AGENTS},architect"
+    log "Added: System architect"
+  fi
+fi
+
+# Research detection (prepend for complex/unclear tasks)
+if [[ "$TASK_LOWER" =~ research|explore|investigate|analyze|study ]] || \
+   [ "$CALCULATED_DIFFICULTY" = "complex" ] || [ "$CALCULATED_DIFFICULTY" = "enterprise" ]; then
+  # Researcher doesn't count against Loop 3 max (research is always valuable)
+  if [ -n "$LOOP3_AGENTS" ]; then
+    LOOP3_AGENTS="researcher,${LOOP3_AGENTS}"
+  else
+    LOOP3_AGENTS="researcher"
+  fi
+  log "Added: Researcher (complex task)"
+fi
+
+# Fill remaining slots with general specialists based on difficulty
+REMAINING_SLOTS=$((LOOP3_MAX - LOOP3_COUNT))
+if [ $REMAINING_SLOTS -gt 0 ] && [ -n "$LOOP3_AGENTS" ]; then
+  log "Filling $REMAINING_SLOTS remaining slots with specialists..."
+
+  # Add coder for general implementation
+  if [ $REMAINING_SLOTS -gt 0 ]; then
+    LOOP3_AGENTS="${LOOP3_AGENTS},coder"
+    REMAINING_SLOTS=$((REMAINING_SLOTS - 1))
+  fi
+
+  # Add performance specialist for enterprise tasks
+  if [ "$CALCULATED_DIFFICULTY" = "enterprise" ] && [ $REMAINING_SLOTS -gt 0 ]; then
+    LOOP3_AGENTS="${LOOP3_AGENTS},perf-analyzer"
+    REMAINING_SLOTS=$((REMAINING_SLOTS - 1))
+  fi
+fi
+
+# Default fallback (if no agents selected)
+if [ -z "$LOOP3_AGENTS" ]; then
+  LOOP3_AGENTS="coder"
+  log "No specific domain detected, using general coder"
+fi
+
+# Ensure Loop 2 count matches suggestion
+CURRENT_LOOP2_COUNT=$(echo "$LOOP2_AGENTS" | tr ',' '\n' | wc -l)
+if [ "$CURRENT_LOOP2_COUNT" -lt "$SUGGESTED_LOOP2" ]; then
+  ADDITIONAL_VALIDATORS=$((SUGGESTED_LOOP2 - CURRENT_LOOP2_COUNT))
+  log "Adding $ADDITIONAL_VALIDATORS additional validators for $CALCULATED_DIFFICULTY difficulty"
+
+  # Add code-quality-validator for comprehensive review
+  if [ $ADDITIONAL_VALIDATORS -gt 0 ]; then
+    LOOP2_AGENTS="${LOOP2_AGENTS},code-quality-validator"
+    ADDITIONAL_VALIDATORS=$((ADDITIONAL_VALIDATORS - 1))
+  fi
+
+  # Add performance benchmarker for enterprise
+  if [ "$CALCULATED_DIFFICULTY" = "enterprise" ] && [ $ADDITIONAL_VALIDATORS -gt 0 ]; then
+    LOOP2_AGENTS="${LOOP2_AGENTS},performance-benchmarker"
+  fi
+fi
+
+# Product Owner (always included)
+PRODUCT_OWNER="product-owner"
+
+# Generate unique task ID
+TASK_ID="cfn-$(echo "$TASK_DESCRIPTION" | tr ' ' '-' | tr '[:upper:]' '[:lower:]' | cut -c1-30)-$(date +%s)"
+
+log "Task ID: $TASK_ID"
+log "Loop 3 agents: $LOOP3_AGENTS"
+log "Loop 2 agents: $LOOP2_AGENTS"
+log "Product Owner: $PRODUCT_OWNER"
+
+##############################################################################
+# Step 2: Invoke Orchestrator in Background
+##############################################################################
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+ORCHESTRATOR="$SCRIPT_DIR/orchestrate-cfn-loop.sh"
+
+if [ ! -f "$ORCHESTRATOR" ]; then
+  echo "Error: Orchestrator not found at $ORCHESTRATOR"
+  exit 1
+fi
+
+log "Launching orchestrator..."
+
+# Build success criteria from task
+SUCCESS_CRITERIA=$(jq -nc \
+  --arg task "$TASK_DESCRIPTION" \
+  '{
+    acceptanceCriteria: ["Implementation complete", "Tests passing", "Code reviewed"],
+    gateThreshold: 0.75,
+    consensusThreshold: 0.90
+  }')
+
+# Launch orchestrator in background
+"$ORCHESTRATOR" \
+  --task-id "$TASK_ID" \
+  --mode "$MODE" \
+  --loop3-agents "$LOOP3_AGENTS" \
+  --loop2-agents "$LOOP2_AGENTS" \
+  --product-owner "$PRODUCT_OWNER" \
+  --max-iterations "$MAX_ITERATIONS" \
+  --epic-context "{\"epicGoal\": \"$TASK_DESCRIPTION\"}" \
+  --success-criteria "$SUCCESS_CRITERIA" \
+  > "/tmp/cfn-exec-${TASK_ID}.log" 2>&1 &
+
+ORCHESTRATOR_PID=$!
+
+log "Orchestrator started (PID: $ORCHESTRATOR_PID)"
+
+# If background mode, return immediately
+if [ "$WAIT_FOR_COMPLETION" = false ]; then
+  if [ "$OUTPUT_FORMAT" = "json" ]; then
+    MONITOR_CMD="redis-cli get \"swarm:${TASK_ID}:status\""
+    jq -nc \
+      --arg task_id "$TASK_ID" \
+      --arg pid "$ORCHESTRATOR_PID" \
+      --arg loop3 "$LOOP3_AGENTS" \
+      --arg loop2 "$LOOP2_AGENTS" \
+      --arg monitor "$MONITOR_CMD" \
+      '{
+        status: "running",
+        task_id: $task_id,
+        orchestrator_pid: ($pid | tonumber),
+        agents: {
+          loop3: ($loop3 | split(",")),
+          loop2: ($loop2 | split(","))
+        },
+        monitor: $monitor
+      }'
+  else
+    echo "Task ID: $TASK_ID"
+    echo "PID: $ORCHESTRATOR_PID"
+    echo "Monitor: redis-cli get 'swarm:${TASK_ID}:status'"
+  fi
+  exit 0
+fi
+
+##############################################################################
+# Step 3: Monitor via Redis (Zero-Token Polling)
+##############################################################################
+
+log "Monitoring orchestrator completion..."
+
+TIMEOUT=21600  # 6 hours max (10 iterations × 60min = 10hrs worst case, but realistically much less)
+ELAPSED=0
+CHECK_INTERVAL=30
+
+while [ $ELAPSED -lt $TIMEOUT ]; do
+  # Check orchestrator status
+  STATUS=$(redis-cli get "swarm:${TASK_ID}:status" 2>/dev/null || echo "")
+
+  log "Status: ${STATUS:-initializing} (elapsed: ${ELAPSED}s)"
+
+  # Check for completion
+  if [ "$STATUS" = "complete" ] || [ "$STATUS" = "cancelled" ] || [ "$STATUS" = "failed" ]; then
+    log "Orchestrator finished: $STATUS"
+    break
+  fi
+
+  # Check if orchestrator process still running
+  if ! kill -0 "$ORCHESTRATOR_PID" 2>/dev/null; then
+    log "Orchestrator process exited"
+    # Check final status
+    STATUS=$(redis-cli get "swarm:${TASK_ID}:status" 2>/dev/null || echo "failed")
+    break
+  fi
+
+  sleep $CHECK_INTERVAL
+  ELAPSED=$((ELAPSED + CHECK_INTERVAL))
+done
+
+if [ $ELAPSED -ge $TIMEOUT ]; then
+  log "Timeout reached, orchestrator still running"
+  STATUS="timeout"
+fi
+
+##############################################################################
+# Step 4: Collect Results from Redis
+##############################################################################
+
+log "Collecting results..."
+
+# Get final consensus
+FINAL_CONSENSUS=$(redis-cli lindex "swarm:${TASK_ID}:metrics:loop2_consensus" 0 2>/dev/null || echo '{"consensus": 0}')
+CONSENSUS_VALUE=$(echo "$FINAL_CONSENSUS" | jq -r '.consensus // 0')
+
+# Get iteration count
+ITERATIONS=$(redis-cli llen "swarm:${TASK_ID}:metrics:iteration_start" 2>/dev/null || echo "0")
+
+# Get deliverables
+DELIVERABLES=$(redis-cli smembers "swarm:${TASK_ID}:deliverables" 2>/dev/null | jq -R -s -c 'split("\n") | map(select(length > 0))')
+
+# Get execution time (first iteration start to last iteration end)
+FIRST_START=$(redis-cli lindex "swarm:${TASK_ID}:metrics:iteration_start" -1 2>/dev/null || echo "0")
+LAST_END=$(redis-cli lindex "swarm:${TASK_ID}:metrics:iteration_duration" 0 2>/dev/null | jq -r '.duration_ms // 0')
+TOTAL_DURATION=$((LAST_END))
+
+##############################################################################
+# Step 5: Return Structured Output
+##############################################################################
+
+if [ "$OUTPUT_FORMAT" = "json" ]; then
+  jq -nc \
+    --arg status "$STATUS" \
+    --arg task "$TASK_DESCRIPTION" \
+    --arg task_id "$TASK_ID" \
+    --arg iterations "$ITERATIONS" \
+    --arg consensus "$CONSENSUS_VALUE" \
+    --arg loop3 "$LOOP3_AGENTS" \
+    --arg loop2 "$LOOP2_AGENTS" \
+    --arg duration "$TOTAL_DURATION" \
+    --argjson deliverables "$DELIVERABLES" \
+    '{
+      status: $status,
+      task: $task,
+      task_id: $task_id,
+      iterations: ($iterations | tonumber),
+      final_consensus: ($consensus | tonumber),
+      agents_selected: {
+        loop3: ($loop3 | split(",")),
+        loop2: ($loop2 | split(","))
+      },
+      deliverables: $deliverables,
+      execution_time_ms: ($duration | tonumber),
+      cost_model: "cli-spawning-95-98pct-savings"
+    }'
+else
+  cat <<EOF
+=== CFN Loop Execution Complete ===
+Status: $STATUS
+Task: $TASK_DESCRIPTION
+Task ID: $TASK_ID
+Iterations: $ITERATIONS
+Final Consensus: $CONSENSUS_VALUE
+Loop 3 Agents: $LOOP3_AGENTS
+Loop 2 Agents: $LOOP2_AGENTS
+Deliverables: $DELIVERABLES
+Execution Time: ${TOTAL_DURATION}ms
+EOF
+fi
+
+# Cleanup
+if [ "$STATUS" != "timeout" ]; then
+  log "Cleaning up Redis keys..."
+  redis-cli --scan --pattern "swarm:${TASK_ID}:*" | xargs -r redis-cli DEL >/dev/null 2>&1 || true
+fi
+
+# Exit with appropriate code
+if [ "$STATUS" = "complete" ]; then
+  exit 0
+elif [ "$STATUS" = "timeout" ]; then
+  exit 124
+else
+  exit 1
+fi

package/.claude/skills/redis-coordination/collect-confidence-scores.sh

@@ -0,0 +1,179 @@
+#!/bin/bash
+#
+# Collect Confidence Scores - Stateless Agent Coordination
+#
+# Extracts confidence scores from multiple agents after they exit.
+# Replaces the 'collect' function from invoke-waiting-mode.sh for stateless architecture.
+#
+# Usage:
+#   ./collect-confidence-scores.sh --task-id <task> --agent-ids <id1,id2,id3> [--min-quorum <0.66|66%|2>] [--namespace swarm]
+#
+# Returns:
+#   Average confidence score (0.0-1.0) to stdout
+#   Verbose messages to stderr
+#
+# Exit Codes:
+#   0 - Success (consensus calculated)
+#   1 - Quorum not met or missing required parameters
+
+set -euo pipefail
+
+# Debug mode (set DEBUG=true for verbose output)
+DEBUG="${DEBUG:-false}"
+
+# Parse arguments
+TASK_ID=""
+AGENT_IDS=""
+MIN_QUORUM=""
+NAMESPACE="swarm"
+
+while [[ $# -gt 0 ]]; do
+    case $1 in
+        --task-id)
+            TASK_ID="$2"
+            shift 2
+            ;;
+        --agent-ids)
+            AGENT_IDS="$2"
+            shift 2
+            ;;
+        --min-quorum)
+            MIN_QUORUM="$2"
+            shift 2
+            ;;
+        --namespace)
+            NAMESPACE="$2"
+            shift 2
+            ;;
+        *)
+            echo "Unknown argument: $1" >&2
+            exit 1
+            ;;
+    esac
+done
+
+# Validate required parameters
+if [ -z "$TASK_ID" ] || [ -z "$AGENT_IDS" ]; then
+    echo "Error: collect-confidence-scores requires --task-id and --agent-ids" >&2
+    echo "" >&2
+    echo "Usage:" >&2
+    echo "  ./collect-confidence-scores.sh --task-id <task> --agent-ids <id1,id2,id3> [--min-quorum <0.66|66%|2>]" >&2
+    exit 1
+fi
+
+# Output verbose messages to stderr
+echo "[Coordinator] Collecting confidence scores from agents..." >&2
+echo "" >&2
+
+# Split agent IDs
+IFS=',' read -ra AGENTS <<< "$AGENT_IDS"
+
+RESULTS=()
+CONFIDENCES=()
+ALL_FEEDBACK=()
+
+for AGENT in "${AGENTS[@]}"; do
+    RESULT_KEY="${NAMESPACE}:${TASK_ID}:${AGENT}:result"
+
+    # Get latest result (non-blocking)
+    RESULT=$(redis-cli LPOP "$RESULT_KEY" 2>/dev/null || echo "")
+
+    if [ -n "$RESULT" ] && [ "$RESULT" != "(nil)" ]; then
+        # Handle both simple numeric format and JSON format
+        # Try to parse as JSON first, fall back to simple number
+        if CONF=$(echo "$RESULT" | jq -r '.confidence' 2>/dev/null) && [ "$CONF" != "null" ]; then
+            # JSON format: {"confidence":0.85,"iteration":1,...}
+            echo "  [$AGENT] Confidence: $CONF" >&2
+            RESULTS+=("$RESULT")
+            CONFIDENCES+=("$CONF")
+
+            # Check if result includes feedback array
+            FEEDBACK=$(echo "$RESULT" | jq -r '.feedback // empty | .[]?' 2>/dev/null || echo "")
+            if [ -n "$FEEDBACK" ]; then
+                echo "  [$AGENT] Feedback provided:" >&2
+                echo "$RESULT" | jq -r '.feedback[]' 2>/dev/null | sed 's/^/    - /' >&2
+
+                # Collect all feedback items
+                while IFS= read -r ITEM; do
+                    [ -n "$ITEM" ] && ALL_FEEDBACK+=("$ITEM")
+                done < <(echo "$RESULT" | jq -r '.feedback[]' 2>/dev/null || echo "")
+            fi
+        elif [[ "$RESULT" =~ ^[0-9]+\.?[0-9]*$ ]]; then
+            # Simple numeric format: "0.85"
+            CONF="$RESULT"
+            echo "  [$AGENT] Confidence: $CONF" >&2
+            CONFIDENCES+=("$CONF")
+        else
+            echo "  [$AGENT] ⚠️  Invalid result format: $RESULT" >&2
+        fi
+    else
+        echo "  [$AGENT] ⚠️  No result found" >&2
+    fi
+done
+
+# Validate quorum if specified
+TOTAL_AGENTS=${#AGENTS[@]}
+RESPONDING_AGENTS=${#CONFIDENCES[@]}
+
+if [ -n "$MIN_QUORUM" ]; then
+    # Parse min-quorum (supports: absolute number, percentage, or decimal)
+    if [[ "$MIN_QUORUM" =~ ^[0-9]+%$ ]]; then
+        # Percentage format: "66%"
+        PCT=${MIN_QUORUM%\%}
+        REQUIRED=$(echo "scale=0; ($TOTAL_AGENTS * $PCT) / 100" | bc)
+    elif [[ "$MIN_QUORUM" =~ ^0\.[0-9]+$ ]]; then
+        # Decimal format: "0.66"
+        REQUIRED=$(echo "scale=0; ($TOTAL_AGENTS * $MIN_QUORUM) / 1" | bc)
+    else
+        # Absolute number format: "2"
+        REQUIRED=$MIN_QUORUM
+    fi
+
+    if [ "$RESPONDING_AGENTS" -lt "$REQUIRED" ]; then
+        echo "" >&2
+        echo "[Coordinator] ❌ Quorum not met" >&2
+        echo "  Required: $REQUIRED agents" >&2
+        echo "  Responding: $RESPONDING_AGENTS agents" >&2
+        exit 1
+    fi
+fi
+
+# Calculate consensus
+if [ ${#CONFIDENCES[@]} -gt 0 ]; then
+    SUM=0
+    for CONF in "${CONFIDENCES[@]}"; do
+        SUM=$(echo "$SUM + $CONF" | bc)
+    done
+    COUNT=${#CONFIDENCES[@]}
+    CONSENSUS=$(echo "scale=2; $SUM / $COUNT" | bc)
+
+    # Ensure leading zero for bc output (handles .87 -> 0.87)
+    if [[ "$CONSENSUS" =~ ^\. ]]; then
+        CONSENSUS="0$CONSENSUS"
+    fi
+
+    echo "" >&2
+    echo "[Coordinator] Average Confidence: $CONSENSUS" >&2
+
+    # Print aggregated feedback if available
+    if [ ${#ALL_FEEDBACK[@]} -gt 0 ]; then
+        echo "" >&2
+        echo "[Coordinator] Aggregated Feedback:" >&2
+        for ITEM in "${ALL_FEEDBACK[@]}"; do
+            echo "  - $ITEM" >&2
+        done
+
+        # Store feedback in Redis for next iteration
+        FEEDBACK_JSON=$(jq -n --argjson items "$(printf '%s\n' "${ALL_FEEDBACK[@]}" | jq -R . | jq -s .)" '$items')
+        FEEDBACK_KEY="${NAMESPACE}:${TASK_ID}:aggregated-feedback"
+        echo "$FEEDBACK_JSON" | redis-cli -x SET "$FEEDBACK_KEY" >/dev/null
+        redis-cli EXPIRE "$FEEDBACK_KEY" 3600 >/dev/null
+    fi
+
+    # Output consensus to stdout (for script callers)
+    echo "$CONSENSUS"
+else
+    echo "" >&2
+    echo "[Coordinator] ❌ No confidence scores collected" >&2
+    exit 1
+fi