Last commit on August 14, 2026·Created on August 4, 2026
coleam00/skills
“A testing framework for auditing and pruning AI instruction layers”
Combined rank
#509
across all skills
Stars
308
7d change unavailable
Forks
98
Watchers
4
Traction scoreGitHub stars can be faked, so popularity alone can be misleading. Traction Score looks for broader signs of real attention, adoption, and active maintenance.
TL;DR
This skill helps developers determine which agent instructions in a repository are still necessary. It runs real tasks across multiple iterations—once with the instruction layer intact and once without—to grade each rule based on how it actually affects the outcome.
By executing these tests in temporary git worktrees, it allows for the safe ablation of context from files like CLAUDE.md or cursor rules without altering the active working tree.
WHO IT'S FOR
AI agent / automation builders
automating web interactions and browser tasks
Platform / DevEx teams
auditing and pruning AI instruction layers
QA / test automation engineers
performing exploratory testing and bug hunts
Backend / infrastructure engineers
managing atomic git commits for AI workflows
Want to install it?
Claude Marketplace · Claude Plugin · Clone to Claude · Manual Copy · NPX Skills AddRepository instructions