CC Insights

Automated archiving and deep analysis of Claude Code interaction history.

Workflow

┌─────────────┐     ┌─────────────┐     ┌─────────────┐     ┌─────────────┐
│   Archive   │ ──► │   Analyze   │ ──► │  Ask User   │ ──► │   Report    │
│   Chats     │     │   Patterns  │     │  (2 phases) │     │   + Summary │
└─────────────┘     └─────────────┘     └─────────────┘     └─────────────┘

Path Variables

Standard Workflow

User: "归档并分析我的CC聊天记录"

1. Run archivechats.py to convert JSONL → Markdown (note the output path as [ARCHIVEROOT])
2. Run analyze_patterns.py to extract patterns
3. Offer analysis options via AskUserQuestion (2 phases)

Post-Archiving Analysis Menu

After archiving, offer options using AskUserQuestion in two phases:

Phase 1: Time Range (Parameter)

Time range is a parameter, not an agent set. It filters all subsequent analysis.

Phase 2: Analysis Dimensions (Agent Set)

Each dimension launches its own set of agents, all receiving the time filter.

AskUserQuestion Template

questions:
  - question: "归档完成！选择时间范围分析？"
    header: "时间范围"
    multiSelect: false
    options:
      - label: "全量分析 (All)"
        description: "分析所有历史交互，总结模式，提出改进建议"
      - label: "近月分析 (30d)"
        description: "分析近30天交互模式和使用习惯"
      - label: "本周分析 (7d)"
        description: "分析近7天交互，适合周度复盘"
  - question: "选择分析维度？"
    header: "分析维度"
    multiSelect: true
    options:
      - label: "多维度深度分析"
        description: "7个领域专项agent并行分析（技能开发、知识管理、内容创作等）"
      - label: "人工干预分析"
        description: "分析首次任务后的用户介入，识别改进点"
      - label: "模式归类"
        description: "识别重复输入，建议创建command/skill/agent"

Required Permissions

Before running this skill, ensure these paths are in ~/.claude/settings.json under permissions.allow:

"permissions": {
  "allow": [
    "Read(~/.claude/skills/cc-insights/**)",
    "Read(~/ClaudeCodeArchive/**)",
    "Read(~/SynologyDrive/ClaudeCodeArchive/**)"
  ]
}

Execution Logic

Based on user selections:

1. Extract time range from Phase 1 selection
2. Read references/timefilterguide.md to get [TIMEFILTER] and [TIMESUMMARY_PREFIX] values
3. For each selected dimension in Phase 2:

- Read the corresponding reference file - Replace [TIMEFILTER] in all agent prompts with the time filter value - Replace [ARCHIVEROOT] with the actual archive path - Launch all agents in parallel using Task tool with subagenttype=Explore - CRITICAL: Prefix each agent prompt with: `` IMPORTANT: Only use Read, Glob, and Grep tools. Do NOT use any browser, web, or network tools. You have permission to read all files under [ARCHIVEROOT]. ``

1. Synthesize results into a single report
2. Apply output naming based on time range (see timefilterguide.md)

Example: User selects "本周分析 (7d)" + "多维度深度分析"

1. [TIME_FILTER] = from the last 7 days only
2. [TIMESUMMARYPREFIX] = 基于近7天数据分析
3. Read references/deepanalysisagents.md
4. For each of the 7 agent prompts, replace [TIME_FILTER] with the filter value
5. Launch 7 agents (all filtered to last 7 days)
6. Output file: CCWeeklyMultiDim_YYYYMMDD.md

Scripts Reference

Scripts are in scripts/ subdirectory relative to this SKILL.md.

python3 <skill_root>/scripts/<script_name>.py [OPTIONS]

archive_chats.py

analyze_patterns.py

generate_insights.py

Additional References

• references/timefilterguide.md - Time parameter values and output naming
• references/deepanalysisagents.md - 7 domain-specific agents
• references/humaninputanalysis.md - 5 intervention analysis agents
• references/pattern_grouping.md - 5 pattern mining agents
• references/workflow.md - Automation setup, cron jobs, best practices