Can this skill reduce API token usage?

Yes, by preventing Claude from going in circles or brute-forcing incorrect solutions, it saves context window space and reduces the number of failed attempts.

When is this skill triggered during a coding session?

It is utilized when Claude detects symptoms of being stuck, such as trying the same approach multiple times, failing tests repeatedly, or losing track of the original task scope.

What happens if the skill's techniques don't resolve the issue?

The framework includes a final fallback step to ask the user for specific clarification, ensuring that time isn't wasted on guesswork when human input is necessary.

How does the When Stuck skill improve Claude's performance?

It provides a structured protocol to stop repetitive failures, forcing the AI to re-evaluate its approach using techniques like root cause tracing, semantic search, or task decomposition.

Stuck Resolution Framework

Name: Stuck Resolution Framework
Author: a-ariff

bya-ariff

0•

生産性とワークフロー

Equips Claude with a strategic framework to identify developmental bottlenecks and pivot to effective recovery techniques when progress stalls.

The When Stuck skill acts as a cognitive guardrail for Claude Code, preventing the AI from falling into repetitive loops or inefficient brute-force attempts during complex development tasks. By monitoring for specific failure signals—such as context window saturation, recurring test failures, or scope creep—this skill enables Claude to automatically consult a dispatch table and transition to a more appropriate problem-solving mode. Whether the situation requires root cause tracing for a 'zombie bug', semantic search for missing logic, or spawning a subagent for a fresh perspective, this skill ensures the agent remains productive and minimizes the need for manual user intervention.

主な機能

010 GitHub stars

02Subagent handoff protocols for gaining a fresh perspective on stubborn problems

03Multi-path dispatch table for handling unclear requirements and missing code

04Strategic task decomposition for overcoming complex or overwhelming tasks

05Structured root cause tracing for persistent and recurring bugs

06Automated recognition of development loops and 'stuck' states

ユースケース

01Breaking out of 'brute force' loops when initial bug fixes fail to resolve test issues

02Clarifying vague project requirements before wasting tokens on incorrect implementations

03Navigating large codebases where the primary agent is struggling to locate specific logic

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add a-ariff/ariff-code-config when-stuck

For use in Claude.ai and ChatGPT

Download Skill