What problem does it solve? AI agents editing Jetpack Compose UI code normally cannot see whether their changes worked, forcing a human to verify every iteration. This Skill gives the agent a closed loop: edit a Kotlin file, trigger a hot reload, capture a device screenshot, and evaluate the result against the design intent without human involvement. ## Core Features & Use Cases - Canonical agent loop: Status check, snapshot session, edit, reload with tier reporting, screenshot evaluation, and loop exit criteria. - Seven verbatim MCP tools: hotswan_get_status, hotswan_reload, hotswan_take_screenshot, hotswan_start_snapshot, hotswan_stop_snapshot, hotswan_select_variant, hotswan_build_and_install. - Snapshot-based rollback: Record every iteration and roll the source back to the best variant when the final state regresses. - Use Case: A developer tells Claude Code to tune a Compose screen until it matches a mockup. The agent edits the composable, reloads in seconds, screenshots the device, compares against the mock, and iterates autonomously. ## Quick Start Ask the agent to tune this Compose screen until it matches the attached mockup using the HotSwan MCP tools, starting with a status check and a snapshot session.
