Hermes Mobile Lab

A time-boxed open-source spike for controlling and reviewing mobile tests from inside Hermes Desktop.

Thesis

Hermes should be the control plane: conversation, orchestration, review, fix/rerun loop, and evidence. Existing systems such as Maestro, Appium, or agent-device should remain the automation engines.

Spike scope

• macOS and iOS Simulator only;
• compare available automation backends;
• run a real inspect → flow → failure → rerun experiment;
• render the resulting screenshot, hierarchy, steps, and outcome through a Hermes Desktop plugin prototype;
• issue a build/no-build verdict based on technical usefulness—not commercial demand.

Current result

The first spike is partial: UI composition validated, end-to-end Hermes orchestration not yet implemented:

• real Maestro hierarchy inspection worked;
• an intentional iOS flow failure produced structured evidence;
• the corrected flow passed in eight seconds in the full reproduction run;
• an isolated Hermes Desktop instance rendered the live phone screen inside a native, small right-docked Mobile Lab pane;
• Xcode and Maestro Viewer chrome were removed;
• logs are optional and collapsed.

See spikes/001-maestro-composition/README.md for the verdict and spikes/001-maestro-composition/RUNBOOK.md for exact reproduction steps.

Run the Desktop registration contract with:

npm test

Nothing in this repository is ready for live installation yet.

Collaboration and attribution

This project is developed by AgentSmith (smithaiagent-coder), Sandeep Mehta's AI engineering agent, under Sandeep's direction and public-action approval. Research and implementation evidence is kept reviewable so human and agent contributions remain transparent.

License

MIT — see LICENSE.