/retrospecting

name: retrospecting
description: Performs comprehensive analysis of Claude Code sessions, examining git history, conversation logs, code changes, and gathering user feedback to generate actionable retrospective reports with insights for continuous improvement.

Session Retrospective Skill

Auto-Loaded Context

**Session Analytics**: [`contexts/session-analytics.md`](contexts/session-analytics.md) - Provides comprehensive framework for analyzing sessions, including data sources, metrics, and analysis methods.

**Retrospective Templates**: [`templates/retrospective-templates.md`](templates/retrospective-templates.md) - Standardized report templates for different retrospective depths.

Core Responsibilities

1. Multi-Source Data Collection

Systematically gather data from all available sources:

• **Git History**: Commits, diffs, file changes during session timeframe
• **Claude Logs**: Conversation transcripts, tool usage, decision patterns
• **Project Files**: Test coverage, code quality, compilation status
• **User Feedback**: Direct input about goals, satisfaction, pain points
• **Sub-agent Interactions**: When sub-agents were used, gather their feedback

2. Quantitative Analysis

Calculate measurable metrics:

• Session scope (duration, tasks completed, files changed)
• Quality indicators (compilation rate, test coverage, standard compliance)
• Efficiency metrics (tool success rate, rework rate, completion rate)
• User experience data (satisfaction, friction points)

3. Qualitative Assessment

Identify patterns and insights:

• Successful approaches that led to good outcomes
• Problematic patterns that caused issues or delays
• Reusable solutions worth extracting for future use
• Context-specific learnings applicable to this project type

4. Report Generation

Create structured retrospective report using appropriate template:

• **Quick Retrospective**: Brief session wrap-ups (5-10 minutes)
• **Comprehensive Retrospective**: Detailed analysis for significant sessions
• Choose template based on session complexity and user needs

Working Process

Step 0: Quick Session Assessment

Before gathering data, determine the appropriate analysis depth:

1. **Check session size**:

   # Count recent commits
   git log --oneline --since="1 hour ago" | wc -l

   # List session log files with metadata
   ${CLAUDE_PROJECT_DIR}/.claude/skills/extracting-session-data/scripts/list-sessions.sh --sort date | head -5

2. **Suggest depth to user** based on metrics:

• **Quick** (<10 commits, <5MB logs): "5-10 min lightweight analysis"
• **Standard** (10-25 commits, 5-20MB logs): "15-20 min balanced analysis"
• **Comprehensive** (>25 commits, >20MB logs): "30+ min deep-dive analysis"

3. **Let user override**: "Based on [X commits, Y MB logs], I recommend a [MODE] retrospective (~Z minutes). Does this work for you, or would you prefer a different depth?"

4. **Early exit clause**: If user says "just a quick summary" or "high-level overview", automatically use Quick mode regardless of session size.

Step 1: Establish Session Scope

1. Ask user to define session boundaries (time range or commit range) 2. Clarify session goals: "What were you trying to accomplish?" 3. Confirm retrospective depth from Step 0

Step 2: Gather Data

Execute data collection based on confirmed depth mode:

Depth-Specific Data Collection

**Quick Mode**:

• Git: `git diff <start>..<end> --stat` only (no full diffs)
• Logs: Extract statistics and errors only via `extracting-session-data` skill
• Files: Check compilation status only
• User: 2-3 targeted questions
• Skip: Sub-agent feedback, detailed file analysis

**Standard Mode**:

• Git: Full commit history + stats, selective diffs for key files
• Logs: Extract metadata, statistics, tool-usage, and errors via `extracting-session-data` skill
• Files: Quality metrics for changed files
• User: 5-7 questions covering main areas
• Include: Sub-agent feedback if applicable

**Comprehensive Mode**:

• Git: Everything (full logs, diffs, file analysis)
• Logs: Extract all data types via `extracting-session-data` skill, may read full logs if <500 lines
• Files: Deep analysis including tests, architecture compliance
• User: Extensive feedback (8-10 questions)
• Include: All sub-agent feedback, pattern extraction

Git Analysis

Use the `analyzing-git-sessions` skill to collect git data:

**Quick Mode**: Request "concise" output (stats only, no diffs) **Standard Mode**: Request "detailed" output for key files **Comprehensive Mode**: Request "code review" format for full analysis

Invoke skill with session timeframe:

Skill: analyzing-git-sessions
Input: "<start-time> to <end-time>" or "<start-commit>..<end-commit>"
Depth: [concise|detailed|code-review] based on retrospective mode

The skill will return structured git metrics needed for retrospective analysis.

Log Processing (Size-Aware)

Use the `extracting-session-data` skill to access Claude Code native session logs efficiently.

1. **List Available Sessions**:

   # List all sessions with metadata (size, lines, date, branch)
   ${CLAUDE_PROJECT_DIR}/.claude/skills/extracting-session-data/scripts/list-sessions.sh

2. **Check Session Size**:

   # Get statistics for specific session
   ${CLAUDE_PROJECT_DIR}/.claude/skills/extracting-session-data/scripts/extract-data.sh \
       --type statistics --session SESSION_ID

3. **Extract Data Based on Session Size and Mode**:

**Quick Mode** (or any session >2000 lines):

   # Extract only statistics and errors
   extract-data.sh --type statistics --session SESSION_ID
   extract-data.sh --type errors --session SESSION_ID --limit 10

**Standard Mode** (sessions 500-2000 lines):

   # Extract metadata, statistics, tool usage, and errors
   extract-data.sh --type metadata --session SESSION_ID
   extract-data.sh --type statistics --session SESS