hotalexnet

Agent Checkpoint — Documentation skill for Claude Code

Documentation community

Repo-local checkpoint and resume skills for coding agents.

How to install Agent Checkpoint

This entry records only its repository, not the path inside it, so there is no exact command to give. Open hotalexnet/agent-checkpoint and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Agent Checkpoint does

Repo-local checkpoint and resume skills for coding agents. Handoffs are plain markdown under .agents/checkpoints/, so Codex can write one and Claude Code or opencode can pick it up later — across sessions, machines, and model switches.

Alternatives in Documentation

会话管理（Session Manager）需求文档（PRD / Markdown） — 目标：对 Codex / Claude Code 的本地会话记录进行可视化管理，并提供“一键复制 / 一键终端恢复”能力 31.7k ★
Qmd Skill — A Codex/Clawd skill definition for qmd (Quick Markdown Search) 673 ★
Cog — 240+ Cognitive architecture for Claude Code -- persistent memory, self-reflection, and foresight via plain-tex 294 ★

README

agent-checkpoint

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT) [![Runtime: Python 3](https://img.shields.io/badge/runtime-python3-blue.svg)](https://www.python.org/) [![State: Repo Local](https://img.shields.io/badge/state-repo--local-2ea44f.svg)](#how-it-works) **English** | [**中文**](./README.zh-CN.md)

Repo-local continuity skills for coding agents. Checkpoints use a shared `agent-handoff/v1` markdown format so Codex/OpenAI CLI, Claude Code, opencode, and other programming agents can write and resume the same repo state.

These two skills turn "I lost the thread" into a recoverable workflow: save the real working lane into the repository itself, then restore it in the next session without depending on fragile external chat memory.

Demo

![Terminal demo of repo-checkpoint and repo-resume](./assets/demo.gif)

What It Does

repo-checkpoint — writes a timestamped markdown handoff under .agents/checkpoints/, including session goal, current state, key chat context, files in play, verification state, next step, and a git snapshot.
repo-resume — restores the latest active lane from the newest checkpoint plus current branch, working tree, and recent commits.

**Why this exists:** Most agents can reread code. What they usually lose is the human context: what the user actually wanted, what was already tried, which constraints mattered, and what the next concrete step should be.

Why It Matters

Repo-local, not session-local — the handoff lives inside the repo, so it survives model switches, browser restarts, shell reconnects, and machine changes.
Cross-agent by design — Codex can write a checkpoint and Claude Code or opencode can resume it later from the same .agents/checkpoints/ directory.
Plain markdown, no lock-in — checkpoints are readable in any editor and reviewable in git.
Fast cold start — resume goes straight to the last known lane instead