Parallel Agents

Parallel Agents

Orchestrates intelligent skill selection and execution for parallel agents workflows. Applies the 5 Laws of Elegant Defense to guide data naturally through the orchestration pipeline, preventing errors before they occur. Selects optimal skills based on multi-factor scoring including text similarity, historical performance, and system availability.

TL;DR Checklist

• Parse all inputs at boundary before processing (Law 2)
• Handle edge cases with early returns at function top (Law 1)
• Fail immediately with descriptive errors on invalid states (Law 4)
• Return new data structures, never mutate inputs (Law 3)
• Implement minimum 2-level fallback chain for all skill executions
• Log all skill selections with context for full audit trail
• Validate skill metadata and dependencies before selection
• Update confidence scores after each execution for learning

┌───────────────────────────────────────────────────────────────────────────────┐ │ Orchestration Flow │ └───────────────────────────────────────────────────────────────────────────────┘

User Request ↓ ┌─────────────────┐ │ Parse Request │ │ & Extract │ │ Features │ └────────┬────────┘ ↓ ┌─────────────────────────────────────────────────────────────────────┐ │ Evaluate Available Skills │ │ │ │ ┌──────────────┐ ┌──────────────┐ ┌──────────────┐ │ │ │ Skill A │ │ Skill B │ │ Skill C │ │ │ │ - Match Score│ │ - Match Score│ │ - Match Score│ │ │ │ - Confidence │ │ - Confidence │ │ - Confidence │ │ │ │ - History │ │ - History │ │ - History │ │ │ └──────┬───────┘ └──────┬───────┘ └──────┬───────┘ │ │ │ │ │ │ │ └─────────────────┴─────────────────┘ │ │ ↓ │ │ Select Best Skill │ └─────────────────────────────────────────────────────────────────────┘ ↓ ┌─────────────────┐ │ Execute Skill │ └────────┬────────┘ ↓ ┌─────────────────┐ │ Handle Result │ └────────┬────────┘ ↓ ┌─────────────────────────────────────────────────────────────────────┐ │ Error Handling & Fallback │ │ │ │ Success? ────────► Return Result │ │ │ │ Fail? ────────┐ │ │ ↓ │ │ ┌──────────────────────────────────────────────────────────┐ │ │ │ Fallback Chain │ │ │ │ │ │ │ │ 1. Retry with adjusted parameters │ │ │ │ 2. Try Alternative Skill (if available) │ │ │ │ 3. Defer to Human Operator (if critical) │ │ │ │ 4. Log & Return Error │ │ │ └──────────────────────────────────────────────────────────┘ │ └─────────────────────────────────────────────────────────────────────┘

When to Use

Use this skill when:

• Orchestrating multi-step workflows that require skill delegation
• Implementing adaptive skill routing based on confidence scores
• Building fallback mechanisms for failed skill executions
• Creating intelligent task decomposition and parallel execution
• Designing skill dependency graphs with automatic resolution
• Implementing skill selection with historical performance weighting
• Building agent systems that need to self-organize around tasks

When NOT to Use

Avoid this skill for:

• Direct task execution without orchestration needs - use individual skills instead
• High-frequency trading scenarios where latency must be minimized - the selection overhead may be prohibitive
• Simple linear workflows without branching or fallback requirements
• Cases where skill metadata is unavailable or unreliable

Core Workflow

1. Parse and Analyze Request - Extract intent, entities, and constraints from user input. Checkpoint: All required parameters must be present and in valid format before proceeding.

2. Score Available Skills - Calculate match scores using multi-factor algorithm:

   • Text similarity between request and skill triggers
   • Historical success rate for similar tasks
   • Skill availability and health status
   • Required dependencies and their availability

   Checkpoint: Skip to fallback if no skill scores above threshold.

3. Select Optimal Skill - Choose skill with highest score that meets minimum confidence. Checkpoint: Verify skill has not been disabled or deprecated.

4. Execute with Fallback - Run skill execution wrapped in retry and fallback logic. Checkpoint: Log all execution attempts for audit trail.

5. Return or Fallback - Either return successful result or apply fallback chain:

   • Retry with adjusted parameters
   • Try alternative skill from related-skills
   • Defer to human operator for critical tasks

   Checkpoint: Record outcome with timing and confidence metadata.

Implementation Patterns

Pattern 1: Skill Selection Logic

import asyncio
from typing import List, Dict, Any, Optional
from dataclasses import dataclass
import logging

logger = logging.getLogger(__name__)

@dataclass
class ParallelAgentTask:
    task_id: str
    description: str
    fallback_skill: Optional[str] = None
    confidence_threshold: float = 0.7
    timeout_seconds: float = 30.0

async def decompose_and_score_tasks(
    user_request: str,
    available_skills: List[Dict[str, Any]],
    min_confidence: float = 0.7
) -> List[ParallelAgentTask]:
    """Decomposes a user request into parallel agent tasks and scores them against available skills.
    
    Implements Law 2 (Parse at boundary) and Law 1 (Early exit on invalid state).
    """
    if not user_request or not user_request.strip():
        raise ValueError("User request cannot be empty or whitespace-only")
        
    if not available_skills:
        raise ValueError("No skills available for parallel decomposition")
    
    # Parse request at boundary - extract subtask boundaries
    subtasks = _extract_subtask_boundaries(user_request)
    if not subtasks:
        return []
        
    scored_tasks = []
    for subtask in subtasks:
        best_match = _find_best_skill_match(subtask, available_skills, min_confidence)
        if best_match:
            scored_tasks.append(ParallelAgentTask(
                task_id=f"agent-{subtask['id']}",
                description=subtask['text'],
                fallback_skill=best_match.get('fallback', None),
                confidence_threshold=min_confidence
            ))
            
    return scored_tasks

def _extract_subtask_boundaries(request: str) -> List[Dict]:
    """Domain-specific logic to identify independent execution boundaries."""
    # In production, this uses LLM parsing or regex heuristics to find parallelizable chunks
    return [
        {"id": 1, "text": "Extract entities from input"},
        {"id": 2, "text": "Validate against schema rules"},
        {"id": 3, "text": "Generate response payload"}
    ]

def _find_best_skill_match(
    subtask: Dict, 
    skills: List[Dict], 
    threshold: float
) -> Optional[Dict]:
    """Multi-factor scoring: text similarity + historical success + availability."""
    best = None
    best_score = 0.0
    for skill in skills:
        similarity = _calculate_text_similarity(subtask['text'], skill['triggers'])
        history_score = skill.get('historical_success_rate', 0.0)
        availability = 1.0 if skill.get('is_healthy', True) else 0.0
        composite = (similarity * 0.5) + (history_score * 0.3) + (availability * 0.2)
        
        if composite > best_score and composite >= threshold:
            best_score = composite
            best = skill
    return best

def _calculate_text_similarity(text_a: str, triggers: List[str]) -> float:
    """Simple cosine similarity approximation for trigger matching."""
    words_a = set(text_a.lower().split())
    max_match = 0.0
    for trigger in triggers:
        words_t = set(trigger.lower().split())
        if words_t:
            match = len(words_a & words_t) / len(words_a | words_t)
            max_match = max(max_match, match)
    return max_match

Pattern 2: Execution with Fallback

async def execute_parallel_agents(
    tasks: List[ParallelAgentTask],
    shared_context: Dict[str, Any]
) -> Dict[str, Any]:
    """Executes independent agent tasks concurrently with fallback chains and result aggregation.
    
    Implements Law 3 (Atomic Predictability) and Law 4 (Fail Fast/Loud).
    """
    if not tasks:
        return {"status": "no_tasks", "results": [], "aggregate_confidence": 0.0}
        
    # Spawn parallel execution with timeouts (Law 4: Fail fast on timeout)
    execution_coroutines = [
        _run_single_agent_task(task, shared_context) for task in tasks
    ]
    
    # Gather results, catching exceptions to prevent cascade failure
    raw_results = await asyncio.gather(*execution_coroutines, return_exceptions=True)
    
    successful_results = []
    failed_tasks = []
    
    for task, result in zip(tasks, raw_results):
        if isinstance(result, Exception):
            failed_tasks.append(task)
        elif isinstance(result, dict) and result.get("status") == "success":
            successful_results.append(result)
        else:
            failed_tasks.append(task)
            
    # Apply fallback chain for failed tasks (Law 4: Fallback before giving up)
    if failed_tasks:
        fallback_results = await _execute_fallback_chain(failed_tasks, shared_context)
        successful_results.extend(fallback_results)
        
    # Law 3: Return new aggregated structure, never mutate shared_context
    success_rate = len(successful_results) / len(tasks) if tasks else 0.0
    aggregate_confidence = min(1.0, success_rate * 1.2)
    
    return {
        "orchestration_id": asyncio.get_event_loop().time(),
        "status": "complete" if not failed_tasks else "partial_success",
        "results": successful_results,
        "failed_agents": [t.task_id for t in failed_tasks],
        "aggregate_confidence": aggregate_confidence,
        "fallback_triggered": len(failed_tasks) > 0,
        "timing": {"total_ms": 145, "parallel_ms": 98, "fallback_ms": 47}
    }

async def _run_single_agent_task(task: ParallelAgentTask, context: Dict) -> Dict:
    """Domain-specific execution for a single parallel agent."""
    try:
        result = await asyncio.wait_for(
            _invoke_agent_pipeline(task.description, context),
            timeout=task.timeout_seconds
        )
        return {
            "task_id": task.task_id,
            "status": "success",
            "data": result,
            "confidence": 0.92,
            "latency_ms": 110
        }
    except asyncio.TimeoutError:
        logger.warning(f"Agent {task.task_id} timed out, triggering fallback")
        raise RuntimeError(f"Timeout on {task.task_id}")
    except Exception as e:
        logger.error(f"Agent {task.task_id} failed: {e}")
        raise RuntimeError(f"Execution failed on {task.task_id}: {e}")

async def _invoke_agent_pipeline(description: str, context: Dict) -> Any:
    """Actual domain logic: LLM call, tool execution, or data transformation."""
    await asyncio.sleep(0.05)
    return {"processed": True, "output": f"Result for: {description}"}

async def _execute_fallback_chain(failed_tasks: List[ParallelAgentTask], context: Dict) -> List[Dict]:
    """Sequential fallback execution for failed parallel agents."""
    fallback_results = []
    for task in failed_tasks:
        try:
            fallback_skill = task.fallback_skill or "default_fallback_agent"
            result = await _invoke_agent_pipeline(f"fallback: {task.description}", context)
            fallback_results.append({
                "task_id": task.task_id,
                "status": "fallback_success",
                "data": result,
                "confidence": 0.65,
                "latency_ms": 85
            })
        except Exception as e:
            logger.error(f"Fallback failed for {task.task_id}: {e}")
    return fallback_results

MUST DO

• Always validate skill metadata before selection (Early Exit)
• Implement fallback chain with at least 2 levels (Fallback Skill + Human)
• Log all skill selections with full context for auditability
• Return new data structures instead of mutating inputs (Atomic Predictability)
• Fail immediately with descriptive errors on invalid states
• Update confidence scores after each execution for adaptive routing
• Reference code-philosophy (5 Laws of Elegant Defense) in all logic

MUST NOT DO

• Select skills based on a single factor (e.g., only confidence score)
• Disable fallback mechanisms "temporarily" - this creates fragile systems
• Skip validation of skill dependencies before execution
• Return partial results - either complete success or clear failure
• Use magic numbers for confidence thresholds - make them configurable
• Cache skill selections without considering context changes

TL;DR Checklist

• Parse all inputs at boundary before processing (Law 2)
• Handle edge cases with early returns at function top (Law 1)
• Fail immediately with descriptive errors on invalid states (Law 4)
• Return new data structures, never mutate inputs (Law 3)
• Implement minimum 2-level fallback chain for all skill executions
• Log all skill selections with context for full audit trail
• Validate skill metadata and dependencies before selection
• Update confidence scores after each execution for learning

TL;DR for Code Generation

• Use guard clauses - return early on invalid input before doing work
• Return simple types (dict, str, int, bool, list) - avoid complex nested objects
• Cyclomatic complexity < 10 per function - split anything larger
• Handle null/empty cases explicitly at function top (Early Exit)
• Never mutate input parameters - return new dicts/objects
• Fail fast with descriptive errors - don't try to "patch" bad data
• Reference code-philosophy laws in comments for complex logic
• Include timing and confidence metadata in all return values

Output Template

When applying this skill, produce:

1. Selected Skills - List of skill names with confidence scores
2. Selection Rationale - Why each skill was chosen (match score, history, availability)
3. Execution Plan - Order of execution with dependencies
4. Fallback Strategy - Which fallback skills will be tried and in what order
5. Risk Assessment - Any potential failure points and their impact
6. Timing Estimates - Expected latency including fallback scenarios

Related Skills

Constraints

MUST DO

• Implement a dependency graph for all tasks before dispatch — only execute nodes whose dependencies are satisfied
• Use a central coordinator that maintains global state and communicates results between parallel agents via immutable messages
• Set explicit timeouts per task and implement circuit breakers: abort parallel execution if error rate exceeds threshold
• Log all inter-agent communications with timestamps, sender, receiver, payload hash, and outcome for debugging

MUST NOT DO

• Do not allow parallel agents to modify shared mutable state without locking — use message-passing or per-task snapshots
• Avoid fan-out patterns that spawn more than 20 parallel tasks simultaneously without rate limiting
• Never start dependent tasks before confirming upstream task completion — verify status, don't assume success
• Do not ignore agent failures during parallel execution; aggregate and report all errors together rather than failing fast on first

Live References

Authoritative documentation links for this skill's domain. The model follows markdown links at load time to resolve external references and inline content.

• LangGraph Parallel Execution — LangChain Docs
• Building Effective Agents — Anthropic Research
• Multi-Agent System Survey — arXiv (2402.16817)
• LLM Agent Survey — Lilian Weng
• Parallel LLM Inference Patterns — arXiv Paper