Capture Learning Command

Manually trigger the learning-analyzer agent to extract learnings from the current session and codify them into skills.

Usage

/capture-learning [optional focus description]

Without arguments: Analyzes the entire session to identify valuable learnings.

With focus description: Guides the analysis to focus on specific topics or patterns.

Examples

/capture-learning

Analyzes full session automatically.

/capture-learning API authentication patterns we discovered

Focuses analysis on API authentication learnings.

/capture-learning React hooks debugging approach

Directs attention to React hooks debugging patterns.

What This Command Does

1. Invokes learning-analyzer agent with current session context
2. Applies quality thresholds to determine if learnings are valuable
3. Smart merge - decides whether to create new skills or update existing ones
4. Creates/updates skill files in configured location (default: .claude/skills/)
5. Stages changes with git (if auto_stage enabled in settings)
6. Provides summary of what was captured

Configuration

Behavior controlled by .claude/session-learner.local.md settings:

• skills_path - Where to save skills (default: .claude/skills/)
• quality_threshold - Skip trivial sessions (default: enabled)
• auto_stage - Automatically git add skills (default: true)
• dry_run - Preview without creating files (default: false)

When to Use

Use this command when:

• You want to capture learnings mid-session (before it ends)
• You've solved a particularly interesting problem
• You've discovered valuable patterns worth codifying
• You want to ensure specific insights aren't lost

Don't use when:

• Session was trivial (typo fixes, simple reads)
• No new patterns or techniques emerged
• Just following standard tutorials

Output

The command provides a summary including:

• Skills created or updated
• Key learnings captured
• File locations
• Git staging status
• Next steps

Integration

This command works with:

• learning-analyzer agent - Does the actual analysis and skill creation
• meta-learning skill - Provides patterns for identifying valuable learnings
• SessionEnd hook - Automatic version runs at session end
• /review-session - Preview mode without creating files

Instructions

When this command is invoked:

1. Check if focus description was provided in the command arguments
2. Invoke the learning-analyzer agent using the Task tool:
   • If no arguments: agent learning-analyzer "Analyze current session and capture learnings"
   • If arguments provided: agent learning-analyzer "Analyze current session focusing on: [arguments]"
3. The agent will handle all analysis, skill creation, and reporting
4. Display the agent's summary output to the user

Important: Always invoke the learning-analyzer agent. Do not attempt to analyze or create skills directly—the agent has the specialized knowledge and tools needed.

Tips

• Be specific with focus descriptions - "database query optimization" is better than "databases"
• Use after solving problems - Capture while the solution is fresh
• Review git diff before committing to see what was generated
• Run /review-session first if unsure what would be captured

Related Commands

• /review-session - Preview learnings without creating files
• /git-status - Check what skills were staged
• /git-diff - Review generated skill content