Conversation Handoff Summary

Generate comprehensive summary for AI agent handoff: $ARGUMENTS

Required Output

This command generates and saves a markdown file. Always:

1. •Analyze the conversation
2. •Create the file using Write tool
3. •Save to: ./handoff-summary-YYYYMMDD-HHMMSS.md

Output Template

# Project Handoff Summary

Generated: [timestamp]

---

## Overview

Brief description of the project and main objectives

## User Context

- Technical background and preferences
- Specific requirements and constraints
- Communication style preferences

## Original Requirements

- Initial problem statement
- Key goals and success criteria

## Completed Work

### Task 1: [Description]

- What was done and why this approach
- Key code changes and files affected

### Task 2: [Description]

...

## Technical Decisions

### Decision 1: [Topic]

- Options considered, chosen approach, rationale

## Current State

- What's working
- Known issues or limitations

## Pending Tasks

- [ ] Task 1
- [ ] Task 2

## Important Warnings

- Critical information to avoid breaking changes
- Security or performance considerations

## Next Steps

Recommended actions for continuing the work

## Related Files

- File 1: Purpose and recent changes
- File 2: Purpose and recent changes

Writing Guidelines

Handoff summaries are high hallucination risk. Apply these constraints:

• •Distinguish completion status: "completed" (verified working), "attempted" (may have issues), "discussed" (not implemented)
• •Quote user's exact words for preferences, constraints, and requirements
• •Only include facts confirmed in conversation -- do not infer unstated decisions or requirements
• •Mark uncertainty: If unsure about a detail, write "[needs confirmation]"
• •Capture the user's actual preferences (e.g., "no emojis", "Korean responses"), not inferred ones

Self-Verification (Required)

Before saving the file, verify:

• • Every "completed" item has evidence in the conversation
• • No tasks listed as done that were only discussed or attempted
• • User preferences accurately reflect stated preferences, not inferred ones
• • Technical decisions include rationale from actual conversation, not fabricated reasoning
• • Pending tasks distinguish user-requested from agent-suggested
• • No information fabricated to "fill" template sections -- leave empty if nothing to report