What is the TOON format used in this skill?

TOON (Token-Oriented Object Notation) is a lightweight syntax designed for uniform arrays and structured data, offering a more token-efficient alternative to JSON or XML for agent communication.

How does this skill prevent context pollution?

It implements delegation rules for high-token tasks (like web scrapes), requiring subagents to return structured summaries within strict token budgets instead of dumping raw data into the main context.

What is the recommended recovery action for malformed subagent output?

The protocol recommends discarding the output entirely and retrying with stricter instructions or escalating the task to a more capable model like Claude 3.5 Sonnet.

When should I use the subagent voting pattern?

Voting should be used for critical or irreversible actions, such as deleting files, deploying code, or performing security assessments, to ensure consensus across multiple model runs.

Subagent Coordination & Recovery

Name: Subagent Coordination & Recovery
Author: rayk

byrayk

•

生産性とワークフロー

Orchestrates and troubleshoots multi-agent workflows to ensure reliable, structured, and accurate outputs for complex tasks.

The coordinate-subagents skill provides a robust framework for managing Claude subagents, specifically focusing on error recovery and high-reliability decision-making. It introduces the TOON (Token-Oriented Object Notation) format for token-efficient data representation and implements parallel voting patterns for critical actions like code deletions or security audits. By establishing clear protocols for model escalation, task decomposition, and context pollution prevention, this skill ensures that complex AI-driven workflows remain stable and cost-effective even when dealing with unpredictable external data or malformed outputs.

主な機能

01Automated red-flag recovery protocols and model escalation strategies

02Atomic task decomposition for improved subagent reliability

03Native support for TOON (Token-Oriented Object Notation) syntax

041 GitHub stars

05Multi-agent parallel voting patterns for high-stakes decisions

06Context pollution prevention for MCP and web-scraping operations

ユースケース

01Extracting structured data from multiple high-token web sources without context bloat

02Debugging and recovering from failed or malformed subagent responses

03Performing consensus-based security assessments for critical code changes

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add rayk/lucid-toolkit coordinate-subagents

For use in Claude.ai and ChatGPT

主な機能

01Automated red-flag recovery protocols and model escalation strategies

02Atomic task decomposition for improved subagent reliability

03Native support for TOON (Token-Oriented Object Notation) syntax

041 GitHub stars

05Multi-agent parallel voting patterns for high-stakes decisions

06Context pollution prevention for MCP and web-scraping operations

ユースケース

01Extracting structured data from multiple high-token web sources without context bloat

02Debugging and recovering from failed or malformed subagent responses

03Performing consensus-based security assessments for critical code changes