Skip to content

rodrigooler/claude-code-prompt-refiner

Repository files navigation

prompt-refiner — sharpen the request before Claude answers

License: MIT Claude Code Plugin PRs welcome

Run /refine and Claude first restates your rough request as a sharp, surgical brief, then answers it.

Half of bad answers come from vague prompts. prompt-refiner adds an on-demand /refine command: it reads your rough request, rewrites it into a tight statement of what you actually want (goal, constraints, success criteria), then proceeds with that understanding. You stay in control because it only runs when you ask for it.

Why on-demand

You don't want every message rewritten. You want it when the request is fuzzy and the cost of a wrong guess is high. So this is a skill you invoke, not a hook that fires on everything.

See it

You type:

/refine make the login better

Claude replies:

🎯 Prompt refined · mode precision
You want to improve the login screen. Likely targets: validation feedback,
error states, and loading state. Assumed scope: UI + client validation, not
the auth backend. Success: clear inline errors, disabled submit while pending,
and a visible failure message. Flagging: confirm if 2FA is in scope.

[then proceeds to do it]

It writes the block in your language and adapts the mode (precision vs creativity) to the task.

Install (2 lines)

/plugin marketplace add rodrigooler/claude-code-prompt-refiner
/plugin install prompt-refiner@claude-code-prompt-refiner

Then use it: /prompt-refiner:refine your rough request.

What it does

  • Restates the request surgically: goal, scope, constraints, expected vs actual, deliverable, success criteria.
  • Picks a mode: precision for bugs/code/ops/facts, creativity for ideation/design/naming/copy.
  • Names hidden assumptions instead of guessing silently.
  • Then answers, without waiting for approval. It only stops to ask if a real ambiguity would change the result.

How it works

One SKILL.md of instructions. No MCP server, no background process, no telemetry. disable-model-invocation keeps it strictly on-demand, so it never fires on its own.

Manual install (without the plugin system)

git clone https://github.com/rodrigooler/claude-code-prompt-refiner
mkdir -p ~/.claude/skills/refine
cp claude-code-prompt-refiner/skills/refine/SKILL.md ~/.claude/skills/refine/SKILL.md

Restart Claude Code and use /refine.

Uninstall

/plugin uninstall prompt-refiner@claude-code-prompt-refiner

Contributing

Issues and PRs welcome — see CONTRIBUTING.md. The whole plugin is one Markdown file.

License

MIT © rodrigooler

About

On-demand prompt refiner for Claude Code — run /refine to turn a rough request into a sharp, surgical brief before Claude acts on it.

Topics

Resources

License

Code of conduct

Contributing

Stars

1 star

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors