Parallelization Patterns

Provide standardized patterns for parallel agent execution, timeout configuration, retry policies, and consensus mechanisms shared across orchestrators and agents. Standard parallel execution capability and strategy patterns Include in agents and commands for consistent parallel execution</use_case> Classification of agents by their parallel execution safety read_only: Safe to run with any other agent (explore, validator) analysis: Safe with other analysis agents (design, database, performance) execution: May modify local state, requires coordination (security, test, devops) orchestration: Manages sub-agents (commands, CLAUDE.md) Standard timeout values by agent type Fast read-only (explore): 180000ms (3 min) Medium analysis (design, database, quality-assurance): 240000ms (4 min) Deep analysis (security, performance, test): 300000ms (5 min) Complex operations (code-quality, devops, git, docs): 300000ms (5 min) Orchestration (commands, CLAUDE.md): 300000ms (5 min) Parallelization pattern for read-only agents (explore, validator) true</parallel_safe> true</read_only> none</modifies_state> 16</max_parallel_agents> 180000</timeout_per_agent> </execution_strategy> all read-only agents </safe_with> Parallelization pattern for analysis agents (design, database, code-quality, performance) true</parallel_safe> true</read_only> none</modifies_state> 16</max_parallel_agents> 240000</timeout_per_agent> </execution_strategy> other analysis agents </safe_with> Parallelization pattern for execution agents (security, test, devops) true</parallel_safe> false</read_only> local</modifies_state> 16</max_parallel_agents> 300000</timeout_per_agent> </execution_strategy> Parallelization pattern for orchestrators (CLAUDE.md, commands) true</parallel_safe> false</read_only> orchestration</modifies_state> 16</max_parallel_agents> 300000</timeout_per_agent> </parallel_groups> </execution_strategy> Standard retry policy for agent execution failures 2</max_retries> Agent timeout Partial results returned Confidence score below 60 </retry_conditions> Use alternative agent from same parallel group </fallback_strategy> </retry_policy> Standard consensus thresholds for multi-agent validation </consensus_thresholds> Standard agent weights for consensus mechanism What type of agent is this? 180000ms 240000ms 300000ms 300000ms 300000ms </decision_tree> What does this agent modify? Use parallelization_readonly Use parallelization_analysis Use parallelization_execution Use parallelization_orchestration </decision_tree> Select appropriate parallelization pattern based on agent capabilities Select timeout based on agent type following decision_tree Group independent agents in parallel_groups for concurrent execution Define depends_on for validation agents that require prior results Use retry_policy for critical operations </best_practices> Read-only agents must use parallelization_readonly or parallelization_analysis Agents that modify state must use parallelization_execution Orchestrators must use parallelization_orchestration Use timeout_selection decision tree for consistent timeout values Apply retry_policy for operations that may fail transiently Use consensus_thresholds for multi-agent validation scenarios Base templates for error escalation, decision criteria, enforcement Output formats, reflection checkpoints, self-evaluation </related_skills> Match parallelization pattern to agent's state modification behavior Use standard timeout tiers from timeout_selection Define parallel_groups with correct dependency chains Running write agents in parallel without coordination Using non-standard timeout values