OhMyOpenCode Files

Engine: senpi 2026.9.18-6

A published install can start its daemon again. Every release from 2026.9.18-4 onward shipped a bundle that could not start a shared host at all: the bundler emitted no host-lifecycle entry, so the deferred import resolved to a chunk nobody wrote, and once that was fixed the launcher spawned the emitted chunk itself - a module, not a program - which returned without ever listening. host ensure answered exited with code 0 before answering get_protocol_info, and the daemon's stderr log was empty because it is truncated on every generation start, so nothing was left to read. Bundled builds now re-enter the CLI through the same internal route compiled binaries use, and the CLI entry comes from the package's declared bin rather than from counting .., which lands on the package root once the module is bundled.

The engine pin moves as one. The senpi version is declared in four manifests plus their peer dependencies and resolutions; advancing only the root resolves two copies at once and produces a type error that reads exactly like a breaking API change but is not. They now move together, and the lockstep test that catches a half-applied bump moves with them.

• 7c4a07cb3 Merge pull request [#8476] from code-yeongyu/release/v5.0.0-beta.76-source-state

• b0b90e531 Merge pull request [#8475] from code-yeongyu/chore/changelog-senpi6

  :::bash npm i -g omo-ai@beta