Bỏ qua tới nội dung
TakoAPI
Quay lại kỹ năng

context-budgeting

bởi sarielwang93

Coding Agents & IDEs
2.1kLượt tải do clawskills.sh công bố
4Sao
0Lượt xem
0 lượt thíchĐăng nhập để thích kỹ năng này
—Chưa có đánh giá
Đánh giá của bạn

Tóm tắt

Manage and optimize OpenClaw context window usage via partitioning, pre-compression checkpointing, and information.

Mô tả

Kỹ năng này làm gì

A framework for managing context window usage in OpenClaw agents. It partitions context into defined zones (goals, history, decisions, background knowledge), checkpoints state before compaction, and runs a cleanup script to recover capacity without restarting the session.

Checkpointing to HOT_MEMORY.md before compaction preserves task state that standard compaction would otherwise discard, allowing the agent to resume accurately.

Khi nào nên dùng

  • Preventing memory loss during long multi-step coding sessions
  • Recovering task state after automatic context compaction
  • Reducing token costs on extended agent workflows
  • Managing multi-megabyte JSON outputs without flooding the context
  • Keeping long-running agents on track past the 80% context threshold

Quy trình ví dụ

ĐẦU VÀO

User asks: Keep my refactoring task going after the context fills up

AGENT
  1. 1

    Check current context usage via /status

  2. 2

    Confirm context exceeds 80% threshold

  3. 3

    Update memory/hot/HOT_MEMORY.md with current status, key decisions made, and next step

  4. 4

    Execute scripts/gc_and_checkpoint.sh to clear raw data and finalize compaction

  5. 5

    Resume the task using the checkpoint as the new working context

ĐẦU RA

Task continues from the last checkpoint without losing prior decisions or progress

Cài đặt

clawhub install sarielwang93/context-budgeting

Chạy lệnh CLI, hoặc dán prompt vào Claude Code, Cursor, Codex hay coding agent bạn đang dùng.

Liên kết