daily-skills-retro

Mine Claude Code transcripts and GitHub review comments to extract reusable coding rules.

1|Updated Apr 3, 2021
One-click install
npx skills add https://github.com/NaoyaMiyagawa/dotfiles --skill daily-skills-retro-naoyamiyagawa
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: daily-skills-retro
Source: https://github.com/NaoyaMiyagawa/dotfiles/tree/main/.claude/scheduled-tasks/daily-skills-retro
Command: npx skills add https://github.com/NaoyaMiyagawa/dotfiles --skill daily-skills-retro-naoyamiyagawa

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Valuable feedback you give on generated code and in review comments is lost after each session; this Skill mines the last 24 hours of that feedback and turns durable corrections into permanent coding rules in your dotfiles skills and AGENTS.md. ## Core Features & Use Cases - Transcript Mining: Runs a bundled shell script with jq programs to extract user text turns and difit inline review comments from Claude Code session transcripts changed in the last 24 hours. - GitHub Review Mining: Pulls your authored pull request review comments via the GitHub API for repositories inferred from recent session activity. - Verified Adoption: Requires verbatim quote verification against transcripts before adopting any rule, then routes each essence to an existing skill, a new skill, or the global AGENTS.md baseline. - Cross-Run Ledger: Maintains a local mining ledger so borderline candidates repeated across days get promoted to real rules. - Use Case: After a day of correcting an AI assistant's Laravel code style and leaving difit review comments, run this Skill to convert those corrections into permanent coding-standard rules. ## Quick Start Run the daily skills retro to mine my feedback from the last 24 hours and adopt durable rules into my dotfiles skills.

Frequently Asked Questions about daily-skills-retro

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

FAQPage Schema
How do I turn my feedback on AI-generated code into permanent rules?

Run this Skill to mine your Claude Code transcripts and GitHub review comments from the last 24 hours. Verified, generalizable corrections are adopted into existing dotfiles skills, new skills, or the global AGENTS.md baseline.

How does the transcript mining script extract user feedback?

The bundled mine.sh script finds jsonl transcripts changed within 24 hours using fd, then runs two jq programs: user-turns.jq extracts user text turns and difit-comments.jq extracts inline review comments left through difit tool results.

Does it use GitHub review comments as a feedback source?

Yes, it queries the GitHub API for pull request comments you authored in the last 24 hours, using repositories inferred from recent transcript project directories rather than hardcoded repo names.

Why are some mined candidates rejected instead of adopted?

Candidates are rejected only when they are task-local to one file or bug, or project-specific to one repository's domain. A candidate whose supporting quote cannot be verified verbatim in the transcript is also rejected to prevent hallucinated rules.

What happens to borderline feedback that only appears once?

Borderline candidates are recorded in a local mining ledger with their verbatim quote and transcript path. If a later run surfaces the same underlying rule, the entry is promoted to a real destination; entries older than 30 days are pruned.