Перейти к содержимому
TakoAPI
Назад к навыкам

hz-context-optimizer

от lidekahdjdhdhsjjs-lang

CLI Utilities
299Число загрузок по данным clawskills.sh
2Просмотры
0 лайковВойдите, чтобы поставить лайк навыку
—Ещё нет оценок
Ваша оценка

Кратко

This skill should be used when the user asks to "compress context", "summarize conversation history", "implement.

Описание

Что делает этот навык

Techniques for compressing agent conversation history without losing critical information. Covers three production-ready approaches: anchored iterative summarization, opaque compression, and regenerative full summaries. The core insight is optimizing tokens-per-task rather than tokens-per-request, since overly aggressive compression forces costly re-fetching.

Targeting tokens-per-task instead of tokens-per-request avoids the hidden cost where compression that saves 0.5% more tokens causes 20% more re-fetching, resulting in higher total token spend.

Когда его использовать

  • Compressing a 100+ message debugging session before hitting context limits
  • Tracking which files were modified across a multi-hour coding agent session
  • Designing summarization triggers for codebases over 5M tokens
  • Evaluating compression quality with probe-based recall tests
  • Reducing total token spend on long-running agentic workflows

Пример сценария

ВВОД

User asks: Compress this 89,000-token debugging session so the agent can continue without losing context

АГЕНТ
  1. 1

    Identify the compression trigger point at 70-80% context utilization

  2. 2

    Define structured summary sections: Session Intent, Files Modified, Decisions Made, Current State, Next Steps

  3. 3

    Summarize the truncated history span into each explicit section

  4. 4

    Merge the new summary into any existing structured summary rather than regenerating from scratch

  5. 5

    Validate retention by running probe questions against the compressed output

ВЫВОД

A structured markdown summary preserving the original error, all modified file paths, root cause, test status, and next steps — compressing 89,000 tokens to roughly 300 words

Установка

clawhub install lidekahdjdhdhsjjs-lang/hz-context-optimizer

Выполните команду CLI или вставьте промпт в Claude Code, Cursor, Codex — любой кодинг-агент, которым пользуетесь.

Ссылки