AgentSkillsCN

manage-context

审计并优化上下文窗口的使用、记忆文件的管理，以及会话效率

SKILL.md
--- frontmatter
name: manage-context
description: Audit and optimize context window usage, memory files, and session efficiency
user-invocable: true
allowed-tools: Read, Write, Grep, Glob, Bash

Context Management

Analyze and optimize the current session's context usage and project memory.

Dynamic Context

  • •Session ID: ${CLAUDE_SESSION_ID}
  • •Memory files: !find .claude/memory -type f -name '*.md' 2>/dev/null | head -20 || echo "No memory files found"
  • •CLAUDE.md size: !wc -l .claude/CLAUDE.md 2>/dev/null || echo "Not found"
  • •Agent memory: !ls .claude/agent-memory/ 2>/dev/null | head -10 || echo "No agent memory"
  • •Rules files: !find .claude/rules -type f -name '*.md' 2>/dev/null | head -10 || echo "No rules found"

Tasks

1. Context Health Check

Analyze context window utilization:

  • •Count loaded files and estimate token consumption
  • •Identify oversized files that could be split or summarized
  • •Detect redundant information loaded from multiple sources
  • •Check for stale @file references in CLAUDE.md

2. CLAUDE.md Audit

Verify CLAUDE.md efficiency:

  • •Measure total size (target: < 2,000 tokens)
  • •Verify all @file references resolve correctly
  • •Identify inline content that should be moved to @references
  • •Ensure IMPORTANT: directives are current and necessary
  • •Check for duplicated information between CLAUDE.md and rules/

3. Memory Optimization

Review and optimize .claude/memory/ contents:

  • •Archive completed task records to a dated archive
  • •Consolidate duplicate or overlapping learnings
  • •Update stale references and outdated patterns
  • •Remove session states older than 7 days
  • •Verify high-value memories (architecture, decisions) are accessible

4. Agent Memory Cleanup

Review .claude/agent-memory/ contents:

  • •Remove memories from agents no longer in use
  • •Consolidate overlapping memories across agents
  • •Verify memory scopes are appropriate (user vs project vs local)

5. Session Recommendations

Based on analysis, provide actionable recommendations:

  • •Whether /clear should be run now
  • •Files to re-read vs. summarize after clearing
  • •Memory entries to archive or consolidate
  • •CLAUDE.md optimizations with specific edits
  • •Rules that could be merged or removed

Output Format

Context Management Report

Current Status

MetricValueStatus
CLAUDE.md[X] lines / ~[Y] tokens✅/⚠️/🔴
Memory files[count] files / [size] total✅/⚠️/🔴
Agent memory[count] entries✅/⚠️/🔴
Rules[count] files✅/⚠️
Estimated context load[LOW/MEDIUM/HIGH/CRITICAL]✅/⚠️/🔴

Issues Found

  1. •[Issue] → [Recommended action]

Actions Taken

  1. •[Action performed and result]

Recommendations

  1. •[Recommendation with rationale]