Subagent History for Claude Code, Codex, Cursor, Grok and Devin

Delegation used to be a spinner and a summary. A child agent would spin up, do real work — read files, run commands, reach a conclusion — and then disappear, leaving you with whatever the parent chose to tell you about it. When the parent's summary was wrong or thin, there was nothing to check it against. Keeping finished children in the list turns a delegation into something you can read after the fact: the actual conversation, the actual tool calls, the actual results. Follow log streams a running child, Earlier logs pages back, Latest jumps to the end, and Copy follow-up for parent hands you a prompt to paste back into the parent conversation so acting on what you found does not mean retyping it. Each child is labelled **Running**, **Done** or **Stopped** with the time of its last activity, and the chip only shows a green dot while something is still running, so a count of children that finished hours ago no longer looks like work in progress. **Send to subagent** appears only while a child is running; a finished one is read-only history. Coverage matters as much as the panel itself. The viewer now opens on Cursor, Grok and Devin terminals too, so the answer to "what did that child actually do?" does not depend on which CLI you happened to delegate from. Each child is bound to the exact parent that spawned it — never the newest session or a same-folder guess — so a delegation cannot be attributed to the wrong terminal, and the names come from the spawn itself: the task title you would recognise, not an id. **Copy ID** hands you a child's id, **Earlier logs** and **Latest** page through the transcript, and the tool-call rows are the receipt when a parent's summary reads thinner than the work.