Перейти к содержимому
TakoAPI
Назад к навыкам

token-saver

от rubenaquispe

Coding Agents & IDEs
4.5kЧисло загрузок по данным clawskills.sh
9Звёзды
1Просмотры
0 лайковВойдите, чтобы поставить лайк навыку
—Ещё нет оценок
Ваша оценка

Кратко

Reduce OpenClaw AI costs with model-aware optimization.

Описание

Что делает этот навык

Reduces AI API costs by compressing workspace files (SOUL.md, USER.md, MEMORY.md, etc.) that are sent with every API call. Detects the active model and adapts compaction thresholds to its actual context window size. Supports 24+ models including Claude, GPT-4, Gemini, and DeepSeek.

Dynamic presets scale compaction thresholds to each model's actual context window, so the same preset works correctly whether you're on GPT-4o's 128K or Gemini's 1M context.

Когда его использовать

  • Cutting API costs on long agent sessions with large workspace files
  • Auditing which workspace files consume the most tokens before a project
  • Setting automatic chat compaction thresholds matched to the active model
  • Compressing MEMORY.md and USER.md into dense key-value format
  • Reverting workspace files to their pre-compression state after testing

Пример сценария

ВВОД

User asks: optimize my workspace for the current model

АГЕНТ
  1. 1

    Detect active model and retrieve its context window from the local registry

  2. 2

    Scan all workspace .md files and calculate token counts

  3. 3

    Apply file-appropriate compression to USER.md and MEMORY.md

  4. 4

    Set compaction threshold to 60% of the detected context window

  5. 5

    Create .backup files for all modified documents

ВЫВОД

Compressed workspace files with auto-backups, compaction threshold set dynamically to model context size

Установка

clawhub install rubenaquispe/token-saver

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

Ссылки