openclaw/tools/delegate-with-healing.sh

#!/usr/bin/env bash
# Delegate task with self-healing retry logic
# Wrapper for manual task delegation med automatic error recovery

set -euo pipefail

REPO_ROOT="/home/alex/clawd"
KANBAN_API="http://192.168.1.220:5003"
LOG_DIR="/home/alex/clawd/logs"
HEALING_LOG="$LOG_DIR/healing-wrapper.log"
MAX_RETRIES=3
HEALING_AGENT="sonnet"

log() {
    echo "[$(date '+%Y-%m-%d %H:%M:%S')] $*" | tee -a "$HEALING_LOG"
}

# Fetch task details from Kanban
get_task() {
    local task_id="$1"
    curl -s "$KANBAN_API/tasks/$task_id"
}

# Execute task with agent
execute_task() {
    local task_id="$1"
    local agent="$2"
    local timeout="$3"
    
    log "🚀 Executing task $task_id with agent $agent (timeout: ${timeout}s)"
    
    # Get task details
    local task=$(get_task "$task_id")
    local title=$(echo "$task" | jq -r '.title // "Untitled"')
    local description=$(echo "$task" | jq -r '.description // ""')
    
    # Build agent prompt
    local prompt="**Task $task_id: $title**

$description

**Instructions:**
- Complete the task as described
- Report progress and results
- Commit any changes to git
- Update task status when done

Work carefully and methodically."
    
    # Execute with agent
    local result=$(cd "$REPO_ROOT" && pnpm -s openclaw agent \
        --local \
        --json \
        --agent "$agent" \
        --timeout "$timeout" \
        --message "$prompt" 2>&1)
    
    echo "$result"
}

# Analyze failure and generate fix
heal_failure() {
    local task_id="$1"
    local agent="$2"
    local failure_output="$3"
    local retry_count="$4"
    
    log "🔧 Healing task $task_id failure (attempt $retry_count/$MAX_RETRIES)"
    
    # Build healing prompt
    local healing_prompt="**Self-Healing Task Recovery**

An OpenClaw agent task failed. Your job is to diagnose and fix it.

**Task ID:** $task_id
**Agent:** $agent
**Retry:** $retry_count/$MAX_RETRIES

**Failure Output:**
\`\`\`
$failure_output
\`\`\`

**Analysis Required:**
1. What went wrong? (root cause)
2. How to fix it? (specific changes)
3. Should we retry? (yes/no + reasoning)

**Actions You Can Take:**
- Edit code/config files
- Install missing dependencies
- Fix syntax errors
- Adjust task parameters
- Recommend different agent/model

Be specific. Make changes if needed. Report back."
    
    # Spawn healing agent
    log "🤖 Spawning healing agent ($HEALING_AGENT)..."
    local healing_result=$(cd "$REPO_ROOT" && pnpm -s openclaw agent \
        --local \
        --json \
        --agent "$HEALING_AGENT" \
        --timeout 300 \
        --message "$healing_prompt" 2>&1)
    
    log "📋 Healing analysis complete"
    
    # Check if healing recommends retry
    if echo "$healing_result" | jq -e '.reply' | grep -qi "retry\|re-run\|try again\|should.*retry"; then
        log "✅ Healing agent recommends retry"
        return 0
    else
        log "❌ Healing agent does not recommend retry"
        return 1
    fi
}

# Main delegation with healing loop
delegate_task() {
    local task_id="$1"
    local agent="${2:-auto}"
    local timeout="${3:-1800}"  # 30 min default
    
    log "📋 Delegating task $task_id to $agent (max retries: $MAX_RETRIES)"
    
    local retry_count=0
    local success=false
    
    while [[ $retry_count -le $MAX_RETRIES ]]; do
        if [[ $retry_count -gt 0 ]]; then
            log "🔄 Retry $retry_count/$MAX_RETRIES for task $task_id"
        fi
        
        # Execute task
        local output=$(execute_task "$task_id" "$agent" "$timeout" || echo "FAILED")
        
        # Check for success
        if echo "$output" | jq -e '.success == true' &>/dev/null; then
            log "✅ Task $task_id completed successfully"
            success=true
            break
        elif echo "$output" | grep -qiE "(completed|done|success|finished)"; then
            log "✅ Task $task_id completed (heuristic match)"
            success=true
            break
        fi
        
        # Task failed - attempt healing
        log "⚠️  Task $task_id failed on attempt $((retry_count + 1))"
        
        if [[ $retry_count -lt $MAX_RETRIES ]]; then
            if heal_failure "$task_id" "$agent" "$output" "$((retry_count + 1))"; then
                retry_count=$((retry_count + 1))
                sleep 5  # Brief pause before retry
                continue
            else
                log "❌ Healing failed - stopping retries"
                break
            fi
        else
            log "❌ Max retries exceeded for task $task_id"
            break
        fi
    done
    
    if [[ $success == true ]]; then
        log "🎉 Task $task_id delegation successful after $retry_count retries"
        return 0
    else
        log "💥 Task $task_id delegation failed after $retry_count retries"
        return 1
    fi
}

# CLI
if [[ $# -lt 1 ]]; then
    echo "Usage: $0 <task-id> [agent] [timeout-seconds]"
    echo ""
    echo "Examples:"
    echo "  $0 335                    # Delegate task 335 (auto-detect agent)"
    echo "  $0 335 gemini-flash      # Use specific agent"
    echo "  $0 335 gemini-flash 3600 # 1 hour timeout"
    exit 1
fi

delegate_task "$@"