Desktop fruit fly fork now follows vibecode markers as scent sources on macOS

The fork adds a scent-tracking layer that watches the filesystem for agent marker files. Each detected marker becomes an odor source whose intensity reflects project openness and marker count. These sources stimulate the fly’s steering neurons (DNa01/DNa02) to bias its walk toward stronger smells. The underlying connectome simulation and macOS sensing are unchanged from the original desktop-fly.

To use the fork a developer clones the repository, runs the supplied build script, and launches the fly application. The build requires macOS 13+ and the Xcode Command Line Tools. The simulation runs as a 1 kHz leaky-integrate-and-fire network, consuming modest CPU while active. Disk scanning for marker files happens in the background and adds little overhead beyond the existing simulation.

If no agent markers are present, the fly behaves exactly as in the upstream version, wandering the desktop without directed movement. Detection is limited to the predefined list of about forty marker files; other project indicators are ignored. The implementation relies on macOS-specific window terrain and permission-free senses, so it does not run on Linux or Windows. Consequently, the scent-following feature stops working outside those environments or when the marker set is empty.

By turning repository metadata into a navigable odor field, the fork shows how project context can be perceived through ambient simulation. Such an approach could inspire future tools that surface code relevance via peripheral cues rather than explicit windows. At present the project remains a demonstrator, with no claim of improving developer productivity or workflow efficiency. Further work would be needed to assess any practical benefit in real-world development settings.