What happens if a checkpoint fails validation?

The coordinator evaluates the drift severity and either issues corrective guidance to the next agent, triggers a phase re-run, or escalates to the user for clarification.

What is goal drift in AI workflows?

Goal drift occurs when sequential AI agents subtly misinterpret instructions over multiple steps, leading the final output to diverge from the original user request.

When should I use the Hierarchical Coordinator skill?

It is best suited for complex tasks involving three or more agents, multi-phase feature builds, or long-running workflows where precision is critical.

Does the coordinator agent write the code itself?

No, the coordinator's sole responsibility is goal alignment. It delegates specialized tasks like coding, testing, and design to other specialist agents.

How does the coordinator agent prevent drift?

It maintains an immutable record of original requirements and validates the output of specialist agents at every phase-gate before allowing the next step to proceed.

Hierarchical Workflow Coordinator

Name: Hierarchical Workflow Coordinator
Author: MadAppGang

byMadAppGang

•

230

•

生产力与工作流

Prevents goal drift in complex multi-agent workflows by implementing a structured coordinator agent and phase-gate validation checkpoints.

The Hierarchical Coordinator skill addresses the fundamental problem of 'goal drift' in long-running AI tasks by introducing a dedicated coordination layer. By establishing an immutable 'Coordinator Context' and enforcing mandatory validation checkpoints between workflow phases—such as architecture, implementation, and testing—the skill ensures that specialist agents remain perfectly aligned with the original user intent. This pattern is essential for complex software engineering projects where sequential handoffs between multiple agents often lead to incremental scope creep or missed requirements.

主要功能

01Structured corrective action loops for minor and major misalignments

02Cross-agent state management using enhanced shared task tracking

03230 GitHub stars

04Immutable requirement tracking via dedicated coordinator-context files

05Phase-gate validation to detect divergence before it accumulates

06Automated drift detection with severity scoring from 'Aligned' to 'Off-Track'

使用场景

01Managing complex system refactors spanning multiple files and architectural layers

02High-precision workflows exceeding 15 minutes where cumulative errors are likely

03Orchestrating multi-phase feature implementations (Plan, Build, Test, Review)

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add madappgang/claude-code hierarchical-coordinator

For use in Claude.ai and ChatGPT

Download Skill