Codex Workflow: When to Use /compact and When to Start a New Session

A community workflow suggests that handoffs are not always necessary for continuing related work. Within the same task, Codex users can often stay in the current session and use context compaction, while unrelated tasks and cross-agent transfers still benefit from a fresh session and a structured handoff.

Published 8/3/2026, 12:13:37 AMLast updated 8/3/2026, 12:13:37 AMEditorial publication

The key decision is whether the work remains part of the same chain of reasoning. If the goal, file scope, and problem remain stable, continuing in the existing session and using /compact can reduce accumulated context. If the work branches into an unrelated task, starting a new session can prevent irrelevant history from affecting later decisions.

For collaboration between agents such as Claude Code and Codex, a handoff can summarize completed work, the current state, important decisions, and recommended next steps. One practical pattern is to have one agent draft and repeatedly review a technical plan, then provide that plan to Codex for implementation.

Acceptance criteria are also critical. Instead of asking an agent to simply “complete the migration,” specify verifiable requirements such as pixel-level UI comparisons, test results, or other evidence. Concrete criteria make it easier for the agent to identify gaps and iterate. Compaction is not lossless, however, so repeated compression, long-running threads, and cross-task reuse can still discard important details.

Source evidence 6

Source review does not guarantee permanent accuracy. Consider the original source, observation time, and later corrections.

Never lose your work between Claude Code sessions - Artem Zhutovartemxtech.substack.com · supporting

Running a retrospective before closing the session. 12 corrections - “this is AI slop” - each one becomes a permanent skill update. Now the agent generates a handoff file. It schedules this auto-continue. I’m not touching anything right now. The agent just types in the next prompt and starts a new session. It tells the other agent - okay, here’s a handoff, follow the instructions, this is continuation from a previous session. This is what was done, this is the current state, and here are the next steps. So now we are in a fresh context window. We started everything clean and we can continue our work this way. [...] The standard way is to do compaction. There is this compact command that frees up context by summarizing the conversation so far. This command is great and it solves this issue, but I have a few problems with that command. One of the problems is I can’t capture learnings right now with that command. How do I transfer the context to another agent? I wan