Review Session Command

Preview what learnings would be captured from the current session without actually creating or modifying any skill files.

Usage

/review-session

This command runs the learning analyzer in preview mode to show what would be captured without writing any files or staging changes.

What This Command Does

1. Analyzes current session for valuable patterns and learnings
2. Identifies relevant skills - shows which would be created or updated
3. Generates preview of skill content without writing files
4. Provides summary of findings without making changes
5. No file modifications - completely safe to run anytime

Output Format

The command provides:

Analysis Summary:

• Overall assessment of session value
• Quality threshold check result
• Number of learnings identified

Skills Preview:

• List of skills that would be created (with names and descriptions)
• List of existing skills that would be updated (with change summaries)
• Draft content preview showing key sections

Key Learnings:

• Bullet points of patterns identified
• Techniques or approaches discovered
• Insights worth capturing

Recommendations:

• Whether capturing is worthwhile
• Suggestions for improving learnings
• Alternative actions if session too trivial

When to Use

Use this command when:

• You want to see what would be captured before committing
• Checking if session has valuable learnings
• Evaluating quality before running /capture-learning
• Curious about what the analyzer would extract
• Before session ends to preview SessionEnd hook behavior

Especially useful:

• Before creating a PR (see what skills would be included)
• After working on complex problems (validate learnings are captured well)
• When unsure if session warrants skill creation

Preview vs. Capture

/review-session	/capture-learning
No files written	Creates/updates skills
No git staging	Stages changes (if enabled)
Safe to run anytime	Modifies repository
Shows preview only	Commits to creating skills
Can run multiple times	Should run when ready

Example Workflow

# 1. Work on problem and solve it
[development work happens]

# 2. Preview what would be captured
/review-session

# 3. Review the output
# - Are the learnings valuable?
# - Are the skills well-structured?
# - Should anything be adjusted?

# 4. If satisfied, capture for real
/capture-learning

# 5. Verify with git
git status
git diff .claude/skills/

Configuration

Uses the same settings as /capture-learning:

• skills_path - Shows where skills would be created
• quality_threshold - Applied during analysis
• Settings from .claude/session-learner.local.md

Note: Always runs in dry-run mode regardless of settings file.

Instructions

When this command is invoked:

1. Invoke the learning-analyzer agent using the Task tool with explicit preview instructions:

   agent learning-analyzer "Analyze current session in PREVIEW MODE. Show what learnings would be captured and what skills would be created/updated, but DO NOT write any files or stage any changes. Provide a detailed summary for review."
   

2. The agent will:

   • Analyze the session
   • Identify learnings
   • Generate skill previews
   • Report findings without modifying files

3. Display the agent's preview output to the user

Important: Ensure the agent knows this is preview mode. The agent should analyze and report but not create files or stage changes.

Output Example

## Session Learning Preview

**Analysis Result:** Found 2 valuable patterns worth capturing

**Quality Check:** ✅ Passed (session had 15 tool calls, 8 minutes of technical work)

**Skills That Would Be Created:**

1. **api-rate-limit-handling** (v1.0.0)
   - Location: .claude/skills/api-rate-limit-handling/SKILL.md
   - Content: Patterns for detecting and handling API rate limits with exponential backoff
   - Sections: Overview, Detection Strategies, Backoff Algorithms, Code Examples

**Skills That Would Be Updated:**

1. **error-handling-patterns** (v1.0.1 → v1.0.2)
   - Location: .claude/skills/error-handling-patterns/SKILL.md
   - Changes: Add section on retry logic with circuit breaker pattern
   - New examples: Circuit breaker implementation from today's work

**Key Learnings:**
- API rate limiting detection pattern (check response headers + status codes)
- Exponential backoff with jitter to avoid thundering herd
- Circuit breaker pattern for failing gracefully after repeated failures

**Recommendation:** ✅ Worth capturing - these are reusable patterns applicable to future API integration work

**Next Steps:**
- Run `/capture-learning` to create these skills
- Or adjust session work and re-run `/review-session`

Tips

• Run before creating PR to see what knowledge will be documented
• Use to validate quality before actually capturing
• Iterate if needed - do more work and review again
• Compare with /capture-learning output to verify consistency

Related Commands

• /capture-learning - Actually create the skills shown in preview
• /git-status - Check repository status (preview doesn't modify)
• /git-diff - View changes (preview has none)