This skill helps developers determine which parts of their agent instructions are actually effective. By running identical tasks in isolated git worktrees—once with the instruction layer intact and once without—it measures the concrete impact of specific rules on agent performance.
It provides a systematic way to reduce context bloat and remove redundant guidelines across various configuration formats, including CLAUDE.md, cursor rules, and copilot instructions.
Follow the documented setup, then try a first task.
Installs: Cole's AI Skills plugin: the complete documented bundle of all 34 skills, managed and read-only. · Claude Code
Enter this complete two-command sequence in a Claude Code session. The first command adds the repository as a plugin marketplace; the second installs the skills bundle from that marketplace.
/plugin marketplace add coleam00/skills /plugin install skills@cole-medin
Uses prime-codebase
Use the prime-codebase skill with no Jira or Confluence arguments to inspect this local repository and give me a concise, easy-to-scan orientation covering its purpose, architecture, technology stack, conventions, current branch, recent activity, and any immediate concerns. Do not modify any files.
Uses piv-review-changes
Use the piv-review-changes skill to review the currently uncommitted changed and new files for real logic, security, performance, and code-quality problems. Verify findings where practical and write the documented review report, but do not fix or commit anything.
Uses system-execution-report
Use the system-execution-report skill to create a structured report for the just-completed feature named [feature-name], using [path-to-plan] as the plan that guided it. Analyze what changed, validation results, challenges, plan divergences, skipped items, and recommendations, then save the report at the documented feature-specific path. Replace [feature-name] and [path-to-plan] with the actual feature name and plan-file path.
Claude
Codex
Cursor
Gemini CLI
Copilot
Windsurf
Amp
OpenCode
Goose
Cline