AgentSkillsCN

Insight Extractor

洞察提取器

SKILL.md

Insight Extractor Skill

Parse the built-in /insights report and extract actionable items into structured, trackable markdown files.

What it does

Reads ~/.claude/usage-data/report.html (generated by /insights) and extracts:

  • •Action Items: CLAUDE.md improvements, unfinished tasks, follow-ups
  • •Useful Prompts & Patterns: Effective interaction strategies and copyable prompts
  • •Technical Learnings: Debugging insights, friction patterns, what worked
  • •Workflow Improvements: Better approaches discovered across sessions
  • •Tool Discoveries: Features to try, new commands, MCP tools
  • •Automation Candidates: Repetitive tasks that could be automated (tagged #automation-candidate, agent-runnable ones tagged #agent-task)

Usage

Auto mode (default)

bash
/insight-extractor

Extracts all categories, saves everything automatically using configured paths.

Interactive mode

bash
/insight-extractor --interactive

Walks through each category with AskUserQuestion -- cherry-pick items, set priorities, choose which automations get task files.

Configure mode

bash
/insight-extractor --configure

Walk through settings via AskUserQuestion. Configures:

  • •Insights folder -- where insight files are saved (default: Research/Insights)
  • •Tasks folder -- where automation task files go (default: Tasks/inbox, or disable)
  • •Daily notes folder -- where daily notes live (default: Daily)
  • •MoC file -- which MoC tracks insights (default: MoC - Insights.md, or disable)
  • •Date format -- YYYYMMDD (compact) or YYYY-MM-DD (ISO)
  • •Machine name -- include in filenames or not
  • •Statistics -- include session stats section or not
  • •Stale threshold -- warn if report older than N days

Settings saved to .claude/skills/insight-extractor/config.json.

Without machine name

bash
/insight-extractor --no-machine

Prerequisites

Run /insights first to generate the report. The extractor reads from:

  • •~/.claude/usage-data/report.html (main report)
  • •~/.claude/usage-data/facets/ (structured JSON data)

Output

Insight file

Research/Insights/YYYYMMDD-insights-{machine}.md with YAML frontmatter and 6 categorized sections.

Daily note link

Appends to today's Daily/YYYYMMDD.md under ## Insights or ## log.

MoC update

Adds entry to MoC - Insights.md under ## Recent Insights.

Task creation

For each automation candidate, creates Tasks/inbox/YYYYMMDD-automate-{slug}.md.

Interactive Mode Details

In --interactive mode, the skill uses AskUserQuestion at each stage:

  1. •Action Items: Include all / review each / skip
  2. •Automation Candidates: Multi-select which ones get task files
  3. •Priority Override: Keep defaults / prioritize CLAUDE.md items / prioritize automation / custom
  4. •Personal Notes: Option to add context before saving

Machine-Specific Behavior

Includes hostname in filename by default (2 machines, different session contexts). MoC groups by date across machines.

When to use

  • •After running /insights: Extract and persist the analysis into your vault
  • •Weekly review: Process insights, review MoC for patterns across days/machines
  • •Automation discovery: Find and track repetitive patterns worth automating

Integration

  • •~/.claude/usage-data/report.html: Source data
  • •MoC - Insights.md: Master index
  • •Daily notes: Quick reference links
  • •Tasks/inbox/: Automation candidates become actionable
  • •MoC - Personal Productivity: Session-as-Memory theme