---
name: checkpoint-summary
description: Create a resumable checkpoint file for another agent to continue work later. Use when user asks to save progress, create handoff notes, checkpoint current state, or leave a resume file for a future agent/session.
---

# Checkpoint Summary

Create a checkpoint file that another agent can read to resume work without reconstructing session history.

## Workflow

1. Identify brief topic for current session. Keep it short enough for filename slug.
2. Resolve skill directory. Bundled script lives in `scripts/` next to this `SKILL.md`.
3. From project root, run `<skill-dir>/scripts/create-checkpoint.py "<topic>"` (absolute script path). NEVER `cd` to skill directory — checkpoint must land in project `.tmp/`, not skill directory. Script defaults to `<git-toplevel>/.tmp/checkpoints`; pass `--output-dir` only to override.
4. Read script output to get created checkpoint path.
5. Fill checkpoint file with current state before ending turn.
6. Tell user exact checkpoint path in final report.

## File Rules

- Store checkpoints under project `.tmp/checkpoints/`.
- Filename MUST include:
  - timestamp
  - unique id
  - brief topic slug
- Do not overwrite prior checkpoints.
- Prefer one checkpoint per meaningful handoff point.

## Checkpoint Contents

Write concise, high-signal content. Another agent should understand:

- task goal
- current status
- decisions already made
- files changed or inspected
- commands/tests already run and key outcomes
- blockers, risks, or assumptions
- exact next steps

## Template Guidance

Populate these sections:

- `Topic`
- `Goal`
- `Current State`
- `Decisions`
- `Files`
- `Commands`
- `Open Questions`
- `Next Steps`

## Rules

- Prefer facts over narrative.
- Include enough context to resume without rereading full chat.
- Do not dump large logs. Summarize outcomes.
- Use repo-relative file paths in file lists when practical.
- If no code changed yet, say so explicitly.