tune

Name: tune
Availability: InStock
Author: duthaho

Mine Claude Code transcripts to rank repeated corrections and propose harness improvements.

8|1|Updated Jul 2, 2026

Search Tags:#workflow automation #claude code #rule refinement #transcript mining #harness retro #permission tuning

Ready to run

npx skills add https://github.com/duthaho/skillhub --skill tune-duthaho

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: tune
Source: https://github.com/duthaho/skillhub/tree/main/.claude/skills/tune
Command: npx skills add https://github.com/duthaho/skillhub --skill tune-duthaho

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill identifies repeated failures across Claude Code sessions and turns them into the smallest effective harness change, so the same mistake stops happening again.

Core Features & Use Cases

Session mining: Reviews recent transcripts for repeated corrections, permission friction, ignored rules, and re-derived procedures.
Root-cause ranking: Weighs findings by frequency and disruption, then filters out one-off noise.
Change proposals: Recommends the weakest mechanism that will actually work, such as a CLAUDE.md rule, allowlist entry, script, hook, new skill, or deletion.
Evidence-driven retro: Requires quoted transcript evidence and stores accepted or rejected findings for future runs.
Use case: Use this when Claude keeps repeating the same mistake, when a workflow is being re-typed every session, or when you want to tighten your project harness without bloating it.

Quick Start

Ask the skill to audit your current Claude Code project for repeated mistakes and propose the highest-impact harness fix first.

Frequently Asked Questions about tune

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I stop Claude Code from making the same mistakes every session?▼

To stop recurring agent mistakes, you can mine Claude Code session transcripts to detect repeated corrections and turn them into ranked harness improvements like CLAUDE.md rules or allowlist entries. This process filters one-off noise and applies the smallest effective fix to prevent the error.

What is the best way to tighten my Claude Code project harness?▼

The best way to tighten a project harness is to run a retrospective on session transcripts, identify ignored rules and permission friction, and apply the weakest sufficient change needed. This proposes targeted updates like hooks or scripts without bloating the configuration.

How do I find which Claude Code rules are causing friction in my workflow?▼

You find rules causing friction by transcript mining recent Claude Code sessions to extract evidence quotes of repeated corrections and permission issues. The findings are weighed by frequency and disruption to pinpoint which instructions need tuning or deletion.

Can I automate workflows that I keep re-typing across Claude Code sessions?▼

Yes, you can automate re-typed workflows by detecting re-derived procedures in session transcripts and proposing a harness change. The system recommends the weakest mechanism that works, such as creating a new skill or adding an allowlist entry to automate the recurring task.

When should I prune dead-weight instructions from my CLAUDE.md file?▼

You should prune dead-weight instructions during a project-level retrospective when transcript evidence shows they are ignored or causing friction. The review process identifies these rules and recommends deletion to keep your configuration clean and effective.