本文へスキップ
TakoAPI
スキル一覧に戻る

context-management

marcus-daemon 作

Web & Frontend Development
558ダウンロード数はclawskills.shが公開している値です
0閲覧数
0件のいいねサインインするとこのスキルにいいねできます
—まだ評価がありません
あなたの評価

概要

Manage AI agent context window consumption, prevent compaction death spirals, and enforce sub-agent spawn policies.

説明

このスキルでできること

Monitors AI agent context window consumption and enforces policies for sub-agent spawning, compaction preparation, and post-compaction recovery. Tracks usage thresholds, estimates remaining runway, and profiles session patterns to recommend config adjustments.

Unlike general memory tools, this skill targets context window mechanics specifically — spawn thresholds, compaction checkpoints, and config tuning — preventing the silent quality degradation that happens before hard limits are hit.

使いどころ

  • Checking how much context runway remains mid-session
  • Deciding whether to spawn a sub-agent for a tool-heavy task
  • Writing a structured checkpoint before triggering compaction
  • Recovering working state after compaction or starting a new session
  • Tuning context compression and pruning settings based on work style

ワークフロー例

入力

User asks: What's eating my context and how much runway do I have?

エージェント
  1. 1

    Run session_status to get current window size and usage percentage

  2. 2

    Estimate breakdown across baseline, messages, tool call results, and compaction summaries

  3. 3

    Classify session pattern as tool-heavy, conversational, mixed, or bursty

  4. 4

    Report usage as a range with threshold-based recommendations

出力

Context usage breakdown with runway estimate and a recommendation to spawn sub-agents or compact based on current percentage

インストール

clawhub install marcus-daemon/context-management

CLIコマンドを実行するか、Claude Code、Cursor、Codexなどお使いのコーディングエージェントにプロンプトを貼り付けてください。

リンク