# Restore

> Resume work after a crash or new session with safety constraints

- Skill: `majiayu000/restore-2` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add majiayu000/restore-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/majiayu000/restore-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: majiayu000 (https://skillmd.com/u/majiayu000)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/majiayu000/restore-2

---


When the user invokes /resume, follow these steps in order:

## 1. Check for Stale State
- List team directories: `ls ~/.claude/teams/` — report any existing teams
- List git worktrees: `git worktree list` — report any worktrees beyond the main one
- Check for stale tmux panes with dead agent sessions

## 2. Clean Up (with user confirmation)
- If stale teams exist, ask user before removing: `rm -rf ~/.claude/teams/<stale-team>/`
- If orphaned worktrees exist, ask user before pruning: `git worktree remove <path>`
- Never auto-clean without user confirmation

## 3. Read Last Checkpoint
- Check `.claude/checkpoints/` for the most recent JSON checkpoint file
- If found, display: timestamp, branch, tasks completed, tasks remaining
- If no checkpoints exist, note this and proceed

## 4. Search Cross-Session Memory
- Use claude-mem search to find recent session context for this project
- Display relevant observations from the last session

## 5. Check Linear Status
- List incomplete Linear tickets for the current project
- Show any in-progress tickets that may need resuming

## 6. Report Summary
Before spawning any agents, present a clear summary:
- What was cleaned up
- Last known checkpoint state
- Recommended next action
- HARD LIMIT reminder: max 3 concurrent agents

## 7. Ask User
Ask the user which task to resume or start next. Do NOT proceed autonomously.