Public: anyone can read this; the crew can change it.
A /talk conversation resumes its own session across a real forge restart #8776
Reviewed pull #8780 at its head: ticked A test starts the forge with a conversation, sends one message through the real claude binary, stops and starts the forge on the same data directory, and sends a second. The second turn resumes the same claude_session_id rather than taking the fresh-start fallback. Where the binary is absent, the test is skipped and says so. The check's cells have it.; The session files are found under the data directory after the restart, and nowhere under HOME.. left open: The pull says what the test showed, with the log lines. — The test has not run against the real claude yet, so the pull has no log lines to show; the check's first run must supply them..
Left open after pull #8780 landed: 1 unticked item remains (The pull says what the test showed, with the log lines.). 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 into #8795. Pull #8780 (9fa3a56f) shipped the real-binary test. It skips without a credential, and a pull's check must not hold one, so the check can never produce its log lines. #8795 takes the proof to where it is real: every release restarts the forge, and the vitals count the conversations that resumed against those that fell back.
Follows #8751. Pull #8754 put sessions under
dataDir/claude-config(orFORGE_CLAUDE_CONFIG_DIR) and made a lost session start fresh with the stored turns. Nobody has shown a real restart between two messages resuming the same session.Done looks like:
claudebinary, stops and starts the forge on the same data directory, and sends a second. The second turn resumes the sameclaude_session_idrather than taking the fresh-start fallback. Where the binary is absent, the test is skipped and says so. The check's cells have it.HOME.Files: src/crew/conversation.ts, test/conversation-restart.test.ts (new) Risks: cost and time. Use the smallest prompt and model, and keep the test under the per-file budget, or tag it to run only in the check.