Shared memory for Claude Code and Codex: pull it on demand, don't inject it

On November 12, 2026, DEV Community's AI section published a piece titled "Claude Code Team Memory: How to Share What Agents Learn," originally hosted at getlockstep.dev. The article bases its walkthrough on Anthropic's official documentation as checked on November 4, explaining where Claude Code's automatic memory is stored. That memory lives in a local user directory on the machine, and the official documentation states clearly that it is not shared across machines or cloud environments, so a teammate's Agent has no way to see the troubleshooting conclusions another engineer reached earlier.

To illustrate the gap this creates, the article walks through a concrete scenario. On a Tuesday, an engineer spends 3 hours with Claude Code on a flaky checkout test, ruling out the payment mock, finding a race in the session store, and getting halfway through a fix before being pulled away to handle an incident. On Wednesday, a teammate picks up the ticket, but their Claude Code starts from zero and once again suggests checking the payment mock first. The article notes that this case sits behind one of the most specific requests in the Claude Code tracker, a feature proposal calling for shared team memory.

On that basis, the article lays out what teams can share at present, what cannot be shared, and where different kinds of knowledge should be stored. As of now, the matter remains at the stage of a documentation-based summary; Anthropic has not announced any further plans for a team-shared memory capability.