How does hybrid routing improve performance?

It uses fast file-pattern matching for low-risk tasks to minimize latency, only escalating to resource-heavy reversibility scoring for RED and CRITICAL tasks.

Can this skill prevent data deletion?

Yes, by classifying data-related tasks as CRITICAL, the skill enforces 'war-room' checkpoints and requires human approval before execution.

What are the four risk tiers used by this skill?

Tasks are classified as GREEN (Safe/Trivial), YELLOW (Caution/Module-level), RED (Danger/Cross-module), or CRITICAL (Stop/Irreversible) based on their potential impact.

What happens if a task has no risk tier assigned?

The skill is designed to be backward compatible, defaulting any unclassified tasks to the GREEN tier for safety and stability.

Risk Classification

Name: Risk Classification
Author: athola

byathola

•

214

•

セキュリティとテスト

Classifies task risk levels using a 4-tier model and automated verification gates to ensure safe AI agent execution.

The Risk Classification skill provides a standardized framework for assessing the operational impact of tasks before they are executed by Claude. It utilizes a 4-tier model—ranging from GREEN (safe) to CRITICAL (irreversible)—to determine the necessary level of oversight and verification. By employing a hybrid routing system, it allows low-risk tasks like documentation updates to proceed via fast heuristic matching while automatically escalating dangerous changes, such as database migrations or production deployments, to advanced reversibility scoring and human-in-the-loop checkpoints.

主な機能

01Metadata extensions for task-level risk tracking

02Automated escalation to war-room checkpoints for high-risk tasks

03214 GitHub stars

044-tier risk categorization (Green, Yellow, Red, Critical)

05Hybrid routing with fast heuristic file-pattern matching

06Parallel execution safety matrix and verification gates

ユースケース

01Determining safe parallel execution limits for cross-module refactors

02Automating safety assessments before assigning complex tasks to AI agents

03Enforcing human approval gates for irreversible infrastructure or database changes

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add athola/claude-night-market risk-classification

For use in Claude.ai and ChatGPT

Download Skill