NextClaw v0.59.7
This release fixes availability regressions in conversation persistence protection and improves underlying error diagnostics. The first trigger on the reporting machine in Issue #73 remains under investigation, and the issue stays open.
Summary failures no longer lock saved conversations
When conversation content was saved successfully, a subsequent catalog summary or preview failure is reported separately instead of permanently refusing new input. One unreadable conversation also no longer prevents healthy conversations from loading during startup scanning.
Send again after storage recovers
A genuine save failure still stops the affected run and reports an error. Accepted input that has not executed remains queued in the current process. After correcting disk space, permissions, or file access, send again in the same conversation. NextClaw waits for the old run to exit, checks saved history, and confirms a recovery write before continuing queued and new input.
A failed request that already started executing is not automatically replayed. Queued input lasts only until the current process exits; this release cannot recover content that was never saved. Keep the original session files.
Underlying diagnostic causes are retained
Logs retain the original error chain, system error codes, operation paths, failure stage, and rollback result. Temporarily unavailable log files no longer interrupt conversations, and logging resumes when access returns.
Upgrade the CLI with npm install -g nextclaw@latest. See when conversation history cannot be saved.