What does 'Hurikaeri' mean?

Hurikaeri is the Japanese word for 'retrospective' or 'reflection.' It is a common practice in agile development to look back at a work cycle to identify improvements.

Where are the retrospective results saved?

Results are stored as YAML files in ~/.claude/hurikaeri/ and can optionally be used to update your project's CLAUDE.md or SKILL.md files.

Can Hurikaeri run automatically?

Yes, it includes a stop hook that detects complex or error-prone sessions and automatically suggests running a retrospective to capture learnings.

How does this skill differ from prompt-improver?

While prompt-improver focuses on improving input instructions across many sessions, Hurikaeri focuses on improving the AI's actual behavior and technical decision-making within the current session.

What is 'counterfactual reasoning' in the context of this skill?

It is a process where the AI analyzes 'what it didn't do' by simulating alternative approaches and identifying missed opportunities or risks it overlooked during the task.

Hurikaeri AI Retrospective

Name: Hurikaeri AI Retrospective
Author: sizukutamago

bysizukutamago

0•

생산성 및 워크플로

Conducts session retrospectives using the AI-KPT framework to analyze AI actions and persist project learnings.

Hurikaeri is a specialized Claude Code skill designed to improve AI-driven development through structured session reviews. It utilizes a three-phase pipeline—Trace, Reflect, and Crystallize—to extract data from session logs and git diffs, apply counterfactual reasoning to identify omissions, and save actionable insights. By formalizing the Keep-Problem-Try (KPT) framework, it helps developers and AI agents evolve their collaboration by updating project documentation like CLAUDE.md and SKILL.md with proven patterns and avoided pitfalls based on real session data.

주요 기능

01AI-KPT analysis to categorize successes, problems, and future improvements

02Counterfactual reasoning to detect 'omissions' and missed technical opportunities

03Automated session tracing from JSONL logs, git diffs, and tool execution history

04Standardized YAML report generation for long-term learning persistence

05Project documentation integration for updating CLAUDE.md and SKILL.md files

060 GitHub stars

사용 사례

01Automating the maintenance of project-specific development rules and best practices

02Identifying why an AI agent missed specific edge cases or alternative architectural approaches

03Post-task analysis to extract reusable patterns from a complex coding session

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add sizukutamago/dev-tools-plugin hurikaeri

For use in Claude.ai and ChatGPT

Download Skill