Edit, Restore, or Handoff: The Three Reset Moves Every Agent User Needs
Three things break an agent session: a wrong turn early, a conversation that drifts, and a thread that grew too big to be useful. Amp gives you three different moves to fix each one. Picking the wrong move wastes time. Picking the right one saves the session.
Edit when the wrong turn is small
You typed a sloppy instruction two messages back. The agent took it literally. The last response is wrong because the instruction was wrong.
Click the message, fix the wording, submit. The context window, the short-term memory the agent works from, resets to that point. Everything after the edited message disappears. The agent runs again from the corrected instruction.
This is the lightest move. Use it when the agent did exactly what you asked and the problem is that you asked the wrong thing.
What it does not do: it does not preserve anything you learned after the wrong turn. If the agent produced something useful in response to the bad instruction, that output is gone. Edit is a rewind, not a patch.
Restore when the conversation derailed
Three messages ago the agent started down a path that does not work. Two messages ago you tried to redirect. One message ago it doubled down. The thread is contaminated.
Hover over the message that started the problem, click Restore. The thread rewinds to before that message. Everything from that point onward leaves the context window.
Restore is heavier than Edit. It throws away real work. Use it when keeping the work would corrupt the next output more than starting fresh would.
The pattern: if you find yourself typing "no, that is not what I meant" more than once, you are past Edit territory. Restore.
Handoff when the thread is too big
You have a thread that produced real results over an hour. Now you want to build on it, but the context is full and quality is dropping. You do not want to lose what you built. You do not want to keep going in this thread either.
Handoff is the move. You tell the agent the goal of the next step. A second model reads the current thread and extracts what is relevant. A fresh thread opens with just that extracted context. You continue in a clean window without starting from zero.
Use Handoff when the work is not done but the thread is. The signal is when you keep re-explaining the same context the agent should already have.
Reference threads without merging them
Sometimes you do not need a handoff. You need one specific thing from a past thread: the SQL query that worked, the API response shape, the exact error message.
Reference the thread by URL or ID. The agent pulls just the relevant fragment into the current context. You skip the handoff entirely.
This is the surgical move. Use it when you know what you need and you know where it lives.
Decision matrix
Edit when: small wording mistake, agent did what you asked but you asked wrong.
Restore when: agent took a path you cannot redirect from in one message.
Handoff when: thread is too long but the work is worth keeping.
Reference when: you need one thing from another thread without the rest.
What none of these fix
No context move fixes an agent that does not understand the project. Edit, Restore, Handoff, and Reference all assume the underlying instructions are sound. If your system prompt or AGENTS.md is the problem, no amount of context surgery will save the session. Fix those first.
Recommended for you
- ContextClaude Code
The MCP You Connected Is Invisible to Your Spawned Agent
MCPs you wire into your main session do not cross into spawned sub-agents. The agent reports Linear unavailable and invents its scope from prompt text.
- ContextClaude Code
Stop Describing Your Codebase. Show It to the Agent.
You described your project in English. The agent produced generic code that matches none of it. Stop describing. Mention the file. Run the command.
- ContextClaude Code
Getting Started with Claude Cowork: A Simple Setup Guide
A simple setup guide for Claude Cowork. Connect your files, tools, and preferences so Claude can do real work alongside you.
- ContextClaude Code
Why Your AI Agent Ignores Your Team's Conventions
You typed the rule in chat three times and the agent still broke it. Chat is for the current task. Project rules live in AGENTS.md. Yours is empty.
Enjoyed this article?
Subscribe for new articles. No spam. Unsubscribe anytime.