Pick up yesterday's work with more than an AI chat history
A morning-resume workflow for recovering what you read, tried, decided, and left unfinished—not just the last message you sent to an agent.
この記事を日本語で読む →
The chat log often starts after the investigation
Agent conversations preserve requests, decisions, revisions, and review results. They matter. But they are only one part of a work session.
Before sending a prompt, you may already have read documentation, inspected code and logs, tested an idea, and narrowed the options. The final request to the agent does not explain all of that earlier work.
What disappears before the first prompt
Suppose you are debugging authentication. You read a GitHub issue and official docs, try a change in your editor, and find that it fails. Only then do you ask Claude Code to implement a different approach.
The agent's chat history may contain the chosen approach, but not the failed attempt or the source that ruled it out. When you return the next morning, those missing steps are exactly what you need to avoid repeating the investigation.
Screens and browser activity can restore clues
An open file, a terminal error, an issue page, and the app you were using can all help identify the point where work stopped. Browser titles and URLs show the shape of an investigation; screenshots and activity timestamps add sequence and visual context.
These signals need interpretation. Opening a page does not prove you read it, and a screenshot does not explain why you made a decision. Treat them as evidence to inspect, not an automatically correct narrative.
Keep memory with the person doing the work
If you switch among Claude Code, Codex, and Cursor, deciding which agent owns the full history becomes awkward. The work happened in your environment, across all three tools.
Contextberg keeps that context on the user's side and exposes selected parts through MCP. The goal is not to replace agent conversations; it is to connect them with browser, screen, and app activity so the next agent can understand the surrounding work.
The minimum useful morning recap
You do not need every event from yesterday. You need the smallest set that tells you where to begin and what still needs checking.
| Question | Why it matters | Possible evidence |
|---|---|---|
| What was I editing? | Find the restart point | Recent app and agent activity |
| What failed? | Continue the investigation | Terminal view and agent conversation |
| What did I read? | Recover possible decision inputs | Browser history and screenshots |
| What did I decide? | Avoid repeating the same debate | Agent history and memory notes |
| What remains open? | Choose the first action | Activity and an explicit note |
A question to ask in the morning
What did I implement in the last two hours of yesterday's work, where did I stop, and what was I going to do next?
List only the pages and observations that may explain the decision.
Mark anything you cannot confirm, then suggest the first 15 minutes of work in order.