When should I use the escalation skill?

Use it when an agent fails or underperforms on a task, requiring a more capable model tier to resolve specific technical challenges.

Can I skip model tiers if a task is very difficult?

The protocol generally prohibits skipping tiers to maintain cost efficiency, though exceptions are made for tasks pre-identified as requiring Opus-level architectural planning.

What triggers a model escalation?

Escalation is triggered by repeated test failures, persistent linting warnings, incorrect output formats, task timeouts, or manual blocks from a tech-lead agent.

Does this skill help save on API costs?

Yes. By starting tasks with Haiku and only escalating to Sonnet or Opus when necessary, it ensures you only use expensive models for tasks that truly require them.

Model Escalation Manager

Name: Model Escalation Manager
Author: eysenfalk

byeysenfalk

0•

Productivity & Workflow

Manages intelligent Claude model tier transitions to optimize task performance and cost efficiency based on performance signals.

The Escalation skill provides a systematic framework for Claude agents to navigate between model tiers—Haiku, Sonnet, and Opus—ensuring tasks are handled by the most capable model required while minimizing unnecessary resource usage. By defining clear escalation signals such as test failures, schema mismatches, and tech-lead blocks, it enforces a 'never-skip-tiers' rule that promotes cost-effective problem solving. This skill is essential for developers building autonomous agent workflows that require high reliability and automated recovery from performance plateaus.

Key Features

01Standardized Haiku-Sonnet-Opus escalation hierarchy

020 GitHub stars

03Performance tracking and logging via claude-mem

04Strict 'never-skip-tiers' logic for resource optimization

05Integrated diagnostic protocol for capability analysis

06Trigger-based escalation for test failures and linting errors

Use Cases

01Scaling agent intelligence dynamically during complex architectural planning

02Automating recovery when a lightweight model fails to resolve code linting errors

03Optimizing API costs by starting tasks with Haiku and escalating only upon failure

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add eysenfalk/git-review escalation

For use in Claude.ai and ChatGPT

Download Skill