orchestration-qa_skill

Overview

This skill provides quality assurance for orchestration workflows by validating that Skills and Subagents follow documented patterns, tracking deviations, and suggesting continuous improvements. It focuses on non-blocking validation: observe, report, and recommend without stopping executions. Configuration is interactive so you only pay token costs for analyses you enable. Reports use clear severity levels (ALERT/WARN/INFO) and track recurring patterns for progressive improvement.

How this skill works

Before a run you can configure which analyses to enable; the skill captures context and sets pre-execution checkpoints. After a Skill or Subagent completes, it compares actual execution against expected workflows and knowledge bases, runs only the conditional analyses you selected, and produces structured findings with recommendations. Findings are stored and aggregated to surface recurring issues and suggest concrete definition or checklist changes.

When to use it

• Configure session analysis categories when starting an orchestration session to control token cost
• Run pre-execution validation before launching a Skill or Subagent to set checkpoints and capture context
• Request a post-execution review after any Skill/Subagent completes to detect deviations and get recommendations
• Enable targeted efficiency analyses (token optimization, tool selection) when optimizing cost or throughput
• Use it for end-of-session pattern tracking to identify recurring issues and suggested improvements

Best practices

• Enable only the analyses you need for the session to minimize token usage
• Always keep Routing Validation enabled for critical checks that prevent status or routing violations
• Run pre-execution validation for complex or cross-domain requests to capture expected checkpoints
• Choose report style (Brief/Standard/Detailed) based on whether you need immediate alerts or full diagnostics
• Act on ALERTs immediately and log WARN/INFO items into backlog for continuous improvement

Example use cases

• Validate a Feature Architect’s PRD extraction to ensure all required sections and templates were applied
• Review a Planning Specialist’s execution graph to find missed parallelization opportunities
• Check a backend Subagent run for tool selection and token efficiency before merging changes
• Detect cross-domain tasks flagged as ALERT and recommend splitting them into domain-isolated tasks
• Track recurring WARN items across a session and produce suggested updates to role-specific checklists