---

name: summary-generator description: Generates session summaries and resume prompts for multi-session work. Use when completing features, before context limits (~50% capacity), or when user says "summary", "wrap up", "save progress", "end session". Creates markdown in docs/summaries/ with completed work, files modified, and restart instructions. allowed-tools: Read, Glob, Grep, Bash(git diff:), Bash(git log:), Bash(git status:*), Write

Session Summary Generator

Overview

This skill creates comprehensive session summaries for complex multi-session work, enabling seamless resumption of tasks. It generates a markdown file in docs/summaries/ with a standardized format.

When to Use

Trigger this skill when:

• User explicitly requests a summary ("generate summary", "wrap up", "save progress")
• Completing a significant feature or refactor
• Conversation context is reaching limits (~50% before auto-compact)
• Before starting a new chat session
• When collaborating with team members on the same feature

Auto-Suggest Triggers

Proactively suggest generating a summary when:

• Multiple files have been modified in the session
• A feature implementation is complete
• The conversation has been long (many exchanges)
• User mentions ending their work session

Output Location

Session summaries are stored in: docs/summaries/YYYY-MM-DD_feature-name.md

Instructions

Step 1: Analyze Current Work

Run these commands to understand what was done:

Copygit status
git diff --stat
git log --oneline -10

Review the conversation history to identify:

• What was accomplished
• Key decisions made
• Files created or modified
• Any remaining tasks

Step 2: Generate Summary File

Create the summary using the template in TEMPLATE.md.

Key sections to include:

1. Overview: Brief description of session focus
2. Completed Work: Bullet points of accomplishments
3. Key Files Modified: Table of files and changes
4. Design Patterns Used: Important architectural decisions
5. Remaining Tasks: What's left to do
6. Resume Prompt: Copy-paste instructions for next session

Step 3: Create Resume Prompt

The resume prompt should include:

• Token optimization directive: Reference to token-optimization.md guidelines (MANDATORY - include at the top)
• Context reference to the summary file
• Specific file paths to review first
• Current status and immediate next steps
• Any blockers or decisions that need user input

IMPORTANT: Always start the resume prompt with:

CopyIMPORTANT: Follow token optimization patterns from `.claude/skills/summary-generator/guidelines/token-optimization.md`:
- Use Grep before Read for searches
- Use Explore agent for multi-file exploration
- Reference this summary instead of re-reading files
- Keep responses concise

Step 4: Analyze Token Usage

Review the conversation for token efficiency opportunities using analyzers/token-analyzer.md and guidelines/token-optimization.md.

Look for:

1. File Reading Patterns

   • Files read multiple times → Suggest caching or using Grep
   • Large files read fully when Grep would suffice
   • Reading generated files (node_modules, build artifacts)

2. Search Inefficiencies

   • Redundant searches → Recommend consolidating queries
   • Overly broad glob patterns
   • Multiple searches that could be combined

3. Response Verbosity

   • Verbose explanations → Note opportunities for conciseness
   • Repeated explanations of same concepts
   • Unnecessary multi-paragraph responses for simple tasks

4. Good Practices to Acknowledge

   • Effective use of Grep before Read
   • Targeted searches with appropriate scope
   • Concise, actionable responses
   • Efficient agent usage

Generate token usage report with:

• Estimated total tokens (use chars/4 approximation)
• Token breakdown by category (file ops, code gen, explanations, searches)
• Efficiency score (0-100) using scoring system from token-analyzer.md
• Top 5 optimization opportunities (prioritized by impact)
• Notable good practices observed

See analyzers/token-analyzer.md for detailed analysis methodology.

Step 5: Analyze Command Accuracy

Review tool calls for accuracy and error patterns using analyzers/command-analyzer.md and guidelines/command-accuracy.md.

Look for:

1. Failed Commands and Causes

   • Path errors (backslashes, wrong case, file not found)
   • Import errors (wrong module path, wrong import style)
   • Type errors (property doesn't exist, type mismatch)
   • Edit errors (string not found, whitespace issues)

2. Error Patterns

   • Categorize by type: path, syntax, permission, logic
   • Identify recurring issues (same mistake multiple times)
   • Note severity: critical, high, medium, low
   • Calculate time wasted on failures and retries

3. Recovery and Improvements

   • How quickly errors were fixed
   • Whether verification prevented errors
   • Improvements from previous sessions
   • Good patterns that prevented errors

Generate command accuracy report with:

• Total commands executed
• Success rate percentage
• Failure breakdown by category
• Top 3 recurring issues with root cause analysis
• Actionable recommendations for prevention
• Improvements observed from past sessions

See analyzers/command-analyzer.md for detailed analysis methodology.

Example Usage

When user says: "Let's wrap up for today"

Response:

1. Analyze git changes and conversation history
2. Create docs/summaries/2025-12-30_enrollment-improvements.md
3. Provide the resume prompt with token optimization directive
4. Suggest: "When context gets long (~40% tokens), start a new chat with this resume prompt"

Example resume prompt output:

CopyResume enrollment improvements session.

IMPORTANT: Follow token optimization patterns from `.claude/skills/summary-generator/guidelines/token-optimization.md`:
- Use Grep before Read for searches
- Use Explore agent for multi-file exploration
- Reference this summary instead of re-reading files
- Keep responses concise

## Context
Previous session completed:
- Added middle name support to enrollment form
- Updated validation for optional middle name field
- Added i18n translations for new field

Session summary: docs/summaries/2025-12-30_enrollment-improvements.md
...

Tips

• Keep summaries focused on a single feature or area
• Include exact file paths for easy navigation
• Note any environmental setup needed (database migrations, etc.)
• Flag any blocking issues or decisions made
• Reference the CLAUDE.md file patterns when applicable