How does the GSD Debugger prevent context window overflow?

The skill spawns isolated subagents for investigation tasks. This 'burns' context in a temporary agent, keeping your main interaction window clean while storing findings in local markdown files.

Does this skill automatically change my code?

No. It focuses on finding the root cause first. Once a cause is confirmed, it presents you with options to fix it automatically, plan a complex fix, or handle it manually.

Can I resume a debugging session if I stop midway?

Yes. The skill tracks active sessions in the .planning/debug/ directory. You can list active sessions and resume them with all previous hypotheses and findings intact.

What information should I provide to start debugging?

The skill will prompt you for expected behavior, actual behavior, specific error messages, the timeline of when the issue started, and reproduction steps.

Which models does the gsd-debugger use?

It uses a model lookup table based on your 'model_profile' setting, typically utilizing Claude 3.5 Sonnet for balanced tasks or Claude 3 Opus for high-quality investigations.

Systematic Debugger (GSD)

Name: Systematic Debugger (GSD)
Author: shoootyou

byshoootyou

•

开发者工具

Diagnoses and resolves complex software issues using a scientific method and isolated subagent investigations.

The GSD Debugger is a high-performance skill designed for deep-dive software investigation within Claude Code. It overcomes context window limitations by spawning specialized 'gsd-debugger' subagents that handle the heavy lifting of reading files, forming hypotheses, and testing, while keeping your main conversation lean. By following a structured process—gathering symptoms (expected vs. actual), managing persistent session states in markdown files, and requiring root cause verification before action—this skill ensures that debugging is methodical, documented, and capable of resuming across context resets.

主要功能

01Isolated subagent execution to preserve main context for user interaction.

02Scientific symptom gathering including reproduction steps and error timelines.

03Support for different model profiles (Opus, Sonnet) based on investigation complexity.

04Checkpoint-driven workflow for human-in-the-loop validation during deep investigations.

0512 GitHub stars

06Persistent session management with state tracking in the .planning/debug/ directory.

使用场景

01Investigating complex logic bugs or race conditions that require reading dozens of files.

02Managing long-running debugging sessions that need to persist across multiple AI interactions.

03Standardizing the debugging process across a development team using spec-driven workflows.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add shoootyou/get-shit-done-multi gsd-debug

For use in Claude.ai and ChatGPT

Download Skill