caveman vs ponytail vs Chisle: why I'd pick Chisle for Claude Code

Описание и применение

Chisle сокращает токены Claude Code до 56% от исходного объёма, обрезая лишний вывод инструментов и сохраняя точность.

Задача

Reduce token usage in Claude Code without sacrificing accuracy.

Результат

Chisle reduces answer length to 56% of baseline, trims tool output by ~46%, and maintains accuracy (no loss in hidden tests).

Рабочий процесс

• Install Chisle with npx chisle
• Run Claude Code with Chisle enabled
• Chisle applies PostToolUse hook to trim large tool output
• Claude generates concise answers
• Agent loop continues with reduced token consumption

Доказательства

Answer length as a percentage of the model with no ruleset (lower is leaner): Claude Code, 26 prompts – caveman 65%, ponytail 58%, Chisle 56%

Источник

Same behaviour. A third of the tokens.

Источник

Tool output was 67.5% of the context: build logs, test runs, giant grep results.

Источник

Chisle also has a hook that trims big tool output before Claude reads it: strips color codes and collapses repeated lines…

Источник