Pick up work in a different agent

Yesterday you were in Claude Code, half way through a task, with a plan declared and three of six tasks done. Today you have Codex open. The session, the plan and the decisions are on your workspace, and Codex has the same kcap tools Claude Code had. The pair is only an example: any two of the coding agents kcap supports work the same way, in either direction

Claude Code stops mid-task, kcap has the session, and a different agent, Codex, picks it up with kcap recap and finishes. Everything below stands on its own

When would I use this

You hit a usage limit. Claude Code stops you at the plan ceiling mid-task; open Codex, recap the session, and finish without waiting for the window to reset
One agent is down. An outage or a broken update on one vendor should not stop the work; another agent picks up from the stored session
A different agent is better at the next step. You planned and explored in one agent and want another’s model or tooling for the implementation or the tests
You moved machines. The session ran on your workstation in one agent; you are on a laptop with a different one installed
It is simply tomorrow. Yesterday’s agent window is gone, the plan is half done, and you would rather resume than re-explain

1. Find the session

You do not need its id. In the agent you are picking up with, Codex here, in the same repository:

Is there an unfinished plan in this repository? Show me where it stopped

The agent lists the open plans for the repository with each one’s progress, next task and the sessions attached. If you know the session, name it instead:

Recap session acme-billing-rate-limit and tell me what remains

Either way the agent reads the stored plan, the outcome summary and as much of the transcript as it needs, and tells you what was decided, what was done and what is left

2. Continue it

Continue from the next task. Keep the plan updated as you go

The agent attaches this session to the same plan, so progress keeps accruing in one place. Open the original session’s Overview in the web app and you will see the task list ticking on, now under a second session

3. Check what you did not see

A plan is finished when its progress says so, not when the agent thinks so. Before you trust the handover, ask:

Show me the failed tool calls from the earlier session and whether they were resolved

kcap errors <session> gives the same answer in the terminal

What made this work

Both agents have the kcap-sessions and kcap-plans tools, installed by setup
The first session declared its plan, so there was a task list to resume rather than a transcript to reread
You can only resume sessions you can see. Your own are always visible; a teammate’s needs to be shared, which is the next walkthrough