Display Name
token-diet
Category
Tooling
Sub-Category
Tooling: Usage Monitors
Primary Link
https://github.com/KasperChenGH/token-diet
Author Name
KasperChenGH
Author Link
https://github.com/KasperChenGH
License
MIT
Description
A zero-dependency CLI, agent, and skill that measures token usage from real Claude Code transcripts, scores structural waste across eight levers, and restructures the project to reduce it. Includes audit-first PostToolUse output-filter and PreToolUse read-dedup hooks, both off by default.
Validate Claims
Zero install: run npx github:KasperChenGH/token-diet review --dir . in any repository that has a .claude/ setup (CLAUDE.md, commands, agents, or skills). It scores the design A-F against the eight levers and lists per-lever findings in seconds, reads no usage history, and writes nothing.
Specific Task(s)
Have Claude Code run token-diet review and then token-diet estimate on a project, and report the letter grade and the projected token-reduction percentage.
Specific Prompt(s)
Tell Claude Code: "Put this project on a token diet - run token-diet review and estimate, then show me the grade and the projected savings."
Additional Comments
Published v0.9.1 (MIT, zero runtime dependencies, 191 tests, CI on Node 18/20/22). The README has a demo GIF showing the output filter's measured reduction table, an A-F design review, and a forward savings estimate.
Recommendation Checklist
Display Name
token-diet
Category
Tooling
Sub-Category
Tooling: Usage Monitors
Primary Link
https://github.com/KasperChenGH/token-diet
Author Name
KasperChenGH
Author Link
https://github.com/KasperChenGH
License
MIT
Description
A zero-dependency CLI, agent, and skill that measures token usage from real Claude Code transcripts, scores structural waste across eight levers, and restructures the project to reduce it. Includes audit-first PostToolUse output-filter and PreToolUse read-dedup hooks, both off by default.
Validate Claims
Zero install: run
npx github:KasperChenGH/token-diet review --dir .in any repository that has a.claude/setup (CLAUDE.md, commands, agents, or skills). It scores the design A-F against the eight levers and lists per-lever findings in seconds, reads no usage history, and writes nothing.Specific Task(s)
Have Claude Code run
token-diet reviewand thentoken-diet estimateon a project, and report the letter grade and the projected token-reduction percentage.Specific Prompt(s)
Tell Claude Code: "Put this project on a token diet - run token-diet review and estimate, then show me the grade and the projected savings."
Additional Comments
Published v0.9.1 (MIT, zero runtime dependencies, 191 tests, CI on Node 18/20/22). The README has a demo GIF showing the output filter's measured reduction table, an A-F design review, and a forward savings estimate.
Recommendation Checklist