You: continue later
Agent/CLI:
  - archives any old continuation.md / continuation-fast.md
  - writes the current git snapshot
  - captures recent local user prompts when enabled
  - leaves continuation.md or continuation-fast.md in the repo root

Later:
You: resume from earlier

Agent:
  - reads the handoff file
  - reports pending tasks, known issues, decisions, gotchas, and

Continue Later: handoffs your next agent can actually use

Ever end a Cursor/Claude/Codex session mid-refactor and come back to a model that doesn't know your branch, your last prompts, or what was left to do?

Continue Later is a small open source toolkit: Agent Skills plus a local CLI that writes handoff files in your repo root — git state, optional recent prompts, tasks, gotchas, and run commands. No hosted service; everything stays on your machine.

What you get

• continue-later — structured continuation.md (overview, stack, state, tasks, decisions, gotchas, deploy steps)
• continue-later-fast — quick continuation-fast.md (git + recent prompts; no extra LLM narrative)
• resume-continuation / resume-from-earlier — natural language (and /resume-from-earlier) to pick work back up
• Optional hooks for Cursor, Codex, Gemini, and a path for Claude Code — auto context on matching prompts when you want it

Try it

One-liner install (see repo for env flags and uninstall):

curl -fsSL https://raw.githubusercontent.com/dhruv-anand-aintech/continue-later-skill/main/install.sh | bash

Then in chat: Use "/continue-later" to save context for your next agent, and "/resume-from-earlier" to pick up where the previous agent left off.

For an LLM-free way of dumping context, from a git root, run:

continue-later-fast 

Check out the source at https://github.com/dhruv-anand-aintech/continue-later-skill