Writes React the way your team already agreed to write it.
Eval suite not run yet
Token efficiency
Token Diet
Cuts what Claude reads before it starts, without cutting what it knows.
What it produces
Trigger accuracy
Not measured yet. This skill does not go on sale until its eval suite has been run against it and the number published here.
What we changed
- Measures before it edits, so every cut has a number attached to it.
- Splits long instruction files into a short always-loaded part and references that load only when needed.
- Refuses to cut anything a trigger description depends on, which is the usual way a diet breaks a skill.
- Reports the saving per file rather than one total, so a bad cut is obvious.
- A troubleshooting table for the case where a skill stops firing after a diet.
What it does, on real work
Finding what is expensive
> what is eating my context in this project?
Ranked every always-loaded file by tokens, and named the three worth splitting first.
Splitting a long instruction file
> move the reference material out of this skill
Cut the always-loaded part and moved the rest to references/, with the loading rule written in.
Checking nothing broke
> did the diet change what it triggers on?
Re-ran the trigger set and reported no change, with the prompts it checked.
Requirements
- Claude Code 2.0+
- macOS, Linux or Windows
Related
Reviews a screen against real UI rules, not taste.
Eval suite not run yet