AgentSkillsCN

context-monitor

实时监测上下文令牌的使用情况，并在接近使用上限时发出预警。 当上下文使用量超过阈值（3万、5万、7万令牌）时，系统会自动触发提醒。 同时提供上下文管理建议，以及“/clear”使用指令。 系统会根据令牌使用量自动触发，无需特定关键词。

SKILL.md
--- frontmatter
name: context-monitor
description: |
  Monitors context token usage and warns when approaching limits.
  Automatically activates when context usage exceeds thresholds (30k, 50k, 70k tokens).
  Provides recommendations for context management and /clear usage.

  Auto-activates based on token usage, no specific keywords needed.
allowed-tools: [Read]

Context Usage Monitor

Purpose

Monitor and warn about context usage to prevent context exhaustion.

Usage Thresholds

  • •30k tokens: First warning (Yellow)
  • •50k tokens: Strong warning (Orange)
  • •70k tokens: Critical warning (Red) - Consider /clear

Monitoring Points

  1. •After each agent invocation
  2. •Before complex tasks
  3. •When reading large files

Warning Messages

Yellow (30k)

code
⚠️ Context Usage: ~30k tokens
Consider focusing on current task only.

Orange (50k)

code
🟠 Context Usage: ~50k tokens
Recommend completing current task then /clear.

Red (70k+)

code
🔴 Context Usage: 70k+ tokens
CRITICAL: Use /clear soon to prevent errors.
Save important context before clearing.

Best Practices

  1. •Use subagents to preserve main context
  2. •Clear context between major features
  3. •Avoid reading unnecessary large files
  4. •Use Task tool for complex searches

Context-Saving Commands

Before /clear, save important info:

bash
# Save current branch and status
git status > /tmp/context_save.txt
git diff --staged >> /tmp/context_save.txt

# After /clear, restore context:
cat /tmp/context_save.txt

Skill Version: v1.0 Last Updated: 2025-12-25 Project: career_ios_backend