OmO 5.1.48 tackles stalled turns and nested-agent task recovery

OmO’s 5.1.48 release targets problems that show up when coding-agent work becomes long and nested: stalled turn endings, recovery after interruption and task ancestry that loses its original root. The October 25 notes describe a bounded wait for a stalled turn-end observer and retryable memory recovery.

For nested tasks, the release says in-process and team ancestry now preserve the root and depth. That matters because a depth limit should not quietly reset when work moves between execution paths. The notes also describe cached repository reads, atomic writes and updated recovery behavior in the underlying senpi runtime.

A separate change removes automatic process_crashed telemetry without distinct crash consent. Manual bug reports remain a separate path. This is a specific consent change, not evidence that every form of analytics has been removed.

OmO is the standalone coding and research agent distributed as omo-ai, with roots in Pi and an older OpenCode plugin edition. The release documents bun add -g omo-ai; provider credentials and inference charges remain separate. Its source-available licensing should not be described as unrestricted permissive open source. The fixes are maintainer-reported changes, not an independent reliability benchmark or proof that every interrupted task will recover.