What problem does it solve? Automated tests can pass while real CLI usage still breaks, especially for image generation, callbacks, and file IO. This Skill provides a change-driven manual testing checklist for the mflux repository so regressions in CLI entrypoints and image outputs are caught before merge. ## Core Features & Use Cases - Change-Driven CLI Checks: Run only the manual checks relevant to what changed on your branch, such as basic generation, model saving, low-RAM mode, stepwise outputs, and metadata sidecars. - Human-in-the-Loop Output Review: Point reviewers at final images, stepwise composites, and metadata JSON files for visual confirmation rather than pixel-perfect comparison. - diffusers Reference Comparison: Validate new MLX model ports against a sibling diffusers clone with matched prompts, seeds, steps, and precision settings. - Use Case: After modifying a CLI flag under src/mflux/models, reinstall the local tool with uv, run the affected generation command with a small step count, and visually confirm the output image is not noise. ## Quick Start Ask the AI to manually test the mflux CLI paths changed on the current branch and review the generated output images.
