本文へスキップ
TakoAPI
スキル一覧に戻る

session-health-monitor

assistantheinrich-prog 作

Coding Agents & IDEs
347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

Context window health monitoring for OpenClaw agents — threshold warnings via Telegram, pre-compaction snapshots.

説明

このスキルでできること

Monitors context window usage in OpenClaw agent sessions and sends threshold-based warnings via Telegram. When usage is high or compaction is detected, the agent saves key facts and decisions to a daily memory file. Old memory files are archived automatically to prevent directory clutter.

Context compaction is silent by default; this surfaces usage and compaction events in real time and preserves key facts before they disappear from the agent's memory.

使いどころ

  • Getting Telegram alerts when an agent context window hits 75%
  • Saving decisions and changed files before a long session ends
  • Detecting when context was silently compacted mid-task
  • Auto-archiving old memory snapshots after a configurable number of days
  • Tracking compaction count across a multi-hour automation run

ワークフロー例

入力

User asks: check session health and send me a Telegram update

エージェント
  1. 1

    Runs bash scripts/context-check.sh --json to get current usage percentage and compaction count

  2. 2

    Evaluates result: 63% used, 1 compaction detected, YELLOW threshold triggered

  3. 3

    Extracts 3-5 key facts from the current session (decisions made, files changed, blockers)

  4. 4

    Runs bash scripts/snapshot.sh to write facts to today's memory file

  5. 5

    Appends footer 📊 63% Context Window | 1x compacted to the outgoing Telegram message

出力

Telegram message sent with context health footer; facts saved to memory/YYYY-MM-DD.md

要件

  • Telegram bot token (obtained from BotFather)
  • Telegram chat ID for the target conversation

インストール

clawhub install assistantheinrich-prog/session-health-monitor

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