Does it record sensitive information?

The skill is built with privacy in mind; it encourages redacting secrets and allows you to set privacy flags to prevent the logging of sensitive project details.

How does Prompt Improver collect feedback?

It uses a Stop hook to trigger a script after tasks are completed, evaluating if the interaction meets certain complexity thresholds (like multiple code changes) before recording feedback.

Where is the feedback data stored?

All feedback is stored locally in your ~/.claude/feedback/ directory as YAML files, ensuring your data remains private and under your control.

Can it help me create new Claude Code skills?

Yes, its recommendation engine analyzes patterns in your feedback to suggest when a recurring task pattern should be graduated into a standalone skill.

What does the /improve command do?

The /improve command analyzes your accumulated feedback files to generate a statistical report on common issues and suggest specific edits to your skills or CLAUDE.md file.

Prompt Improver

Name: Prompt Improver
Author: sizukutamago

bysizukutamago

0•

Productivity & Workflow

Automates the collection and analysis of task feedback to continuously refine Claude's instructions, skills, and configuration.

Prompt Improver establishes a self-correction loop for your Claude Code environment by automatically capturing feedback after task completion. It monitors code changes and complex interactions to identify where instructions in CLAUDE.md or specific skills may be ambiguous or incomplete. By providing structured analysis through the /improve command, it offers actionable recommendations for new skills, documentation updates, and pattern refinements, ensuring your AI assistant becomes increasingly accurate and aligned with your workflow over time.

Key Features

01Structured YAML storage for task outcomes, issues, and severity levels.

02Advanced pattern analysis to identify recurring friction points and instruction gaps.

030 GitHub stars

04Interactive improvement suggestions for CLAUDE.md and existing skill sets.

05Automated feedback collection via Stop hooks based on task complexity and code changes.

06Automatic recommendations for splitting or creating new skills based on usage data.

Use Cases

01Identifying when a set of repeated tasks warrants the creation of a specialized new skill.

02Auditing AI performance across complex multi-step development sessions to ensure high reliability.

03Refining project-specific coding standards in CLAUDE.md based on past misunderstandings.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add sizukutamago/dev-tools-plugin prompt-improver

For use in Claude.ai and ChatGPT

Download Skill