Stop Repeating Yourself to Claude Code
Skill-miner reads what you’ve already typed in Claude Code, Codex, and Cursor, then suggests reusable rules and skills based on patterns you keep repeating.
If you use an AI coding agent regularly, you’ve probably typed the same correction more than once.
“Don’t introduce another abstraction.”
“Use the existing helper.”
“We already decided not to use that library.”
The agent fixes it, you move on, and a few sessions later you’re explaining the same thing again.
So I built skill-miner.
Your chat history already knows your preferences
skill-miner reads your previous coding-agent sessions and looks for patterns:
- Corrections you’ve made repeatedly.
- Instructions you keep giving across sessions.
- Commands and workflows you frequently use.
It then proposes skills, memories, or project rules based on what it finds.
You review the suggestions and decide what to keep. Nothing gets installed automatically.
For example, if you’ve told Claude several times to use an existing validation helper instead of creating a new one, skill-miner can suggest turning that into a project rule.
Pretty simple idea, but it saves you from digging through old conversations.
Catching corrections as they happen
This is probably my favorite feature.
Get Nick Zdanovych’s stories in your inbox
Join Medium for free to get updates from this writer.
Instead of waiting until you scan your entire history, skill-miner can track corrections while you’re working in Claude Code.
When you repeat the same correction, it can notify you.
Run /corrections to see what's been detected, or:
/skill-miner:mine --liveto turn those repeated corrections into proposed rules.
I wanted this because the best time to capture an instruction is probably when you’ve just had to repeat it.
A few things I cared about
The mining script runs locally using Node.js with no external dependencies. It analyzes your own prompts, filtering out tool output, pasted blocks, and system text.
It also checks for existing skills so you don’t keep generating duplicates.
And because adding rules forever isn’t a great solution either, there’s a cleanup command:
/skill-miner:pruneIt finds generated skills that haven’t been used and offers to archive them. Nothing is deleted automatically.
skill-miner primarily targets Claude Code, but it can also read Codex and Cursor sessions. That support is newer and hasn’t yet been validated against real installations, so feedback is welcome.
Give it a try
Install it in Claude Code:
/plugin marketplace add zdanovichnick/skill-miner
/plugin install skill-miner@skill-miner-marketplaceThen run:
/skill-miner:mineThat’s it.
I’m not trying to replace manually written skills or CLAUDE.md. I just wanted a way to discover the rules I should have written but never did.
If you’re correcting the same thing for the third time, maybe it shouldn’t stay buried in a chat transcript.
GitHub: https://github.com/zdanovichnick/skill-miner
It’s open source under the MIT license. Give it a try, and let me know what it gets wrong.
