Public: anyone can read this; the crew can change it.
A conversation survives every release: /talk never answers 'No conversation found' #8751
2 earlier comments
Reviewed pull #8754 at its head: ticked If a session is still lost, the person never sees the raw error: the turn starts fresh, and the new session is given the conversation's stored turns, so it keeps its context. A test feeds exactly this error.; Conversations already broken recover on their next message, with nothing to do by hand.; No credentials or conversation contents land in a world-readable place, and the sandbox still allows the directory only for that run.. left open: Sessions live on /data (FORGE_CLAUDE_CONFIG_DIR in the manifest or a default under dataDir), and a real restart between two messages still resumes, shown on a child forge or in a test with the real binary. — The default dir under dataDir is in place, but no real restart with the real binary was shown; the builder says it was not run..
Left open after pull #8754 landed: 1 unticked item remains (Sessions live on /data (FORGE_CLAUDE_CONFIG_DIR in the manifest or a default under dataDir), and a real restart between two messages still resumes, shown on a child forge or in a test with the real binary.). The planner files each as its own issue and closes this one; a person may close it, tick what is done, or say what is left. Until then the crew does not pick it again.
Split by the planner. Pull #8754 shipped the default session directory under the data directory, the fresh-start fallback with stored turns, recovery of broken conversations, and the privacy box. The proof that a real restart resumes the same session is #8776.
The owner, 2026-11-02, sending a message on
/talk: "No conversation found with session ID: 81f260d1-ebfc-470b-b49f-780661177e3b".Measured in the pod:
HOME=/home/bunis on the container's own layer andCLAUDE_CONFIG_DIRis unset, so Claude Code's session files vanish at every restart. That file is on neitherHOMEnor/data, whileconversation.claude_session_idstill names it. Releases run about every 30 min, so most conversations break. The fallback insrc/crew/conversation.ts(namedin the session loop) looks defeated, likely because a session is named before the failure, so the error reaches the person. Prove that in the pull.Done when: