Use the ConvoAI 2D skill for Hands-on 1.

Read .agents/skills/convoai-exploration-2d/SKILL.md and its required references. Read PROJECT.md, then inspect inputs/pfa-option3-concept.png. Use the rightmost panel as the design image; the other panels explain its conceptual origin. Additional reference images/material vocabulary: none.

Design enquiry: How can a group of roofs make arrival and shared outdoor space more legible while keeping a visual connection towards the sea?

Settings: Massing × Divergent, exactly one round. New run; continuation history is not applicable. Preserve the shared ground/platform concept, a recognisable arrival side and the oblique view direction. Explore the roof masses and the shared open space. Treat the diagram as conceptual, with no measured site or engineering claim. Do not redesign the other two explanatory panels or present a three-panel sheet as the new option.

Complete the skill's input manifest from these supplied settings and PROJECT.md. Ask only about a genuinely unresolved required input. Save the session under a new run folder in outputs/exercise-01/.

Show and save Design Analysis, Full Analysis, Design Advice and Full Recommendations before using the image tool. Then generate one concept image using the real image tool available in this desktop chat. Inspect it against the input and explain what improved, what changed unexpectedly and one next design decision. Preserve the input and any failed attempt. Allow at most one diagnosed retry if a critical constraint fails.

In the final response, show the numbered round as analysis/advice, actual image, then reflection, and link the saved transcript and image. If image generation is unavailable, report the exact limitation and stop with “analysis/advice only”; do not claim Hands-on 1 is complete or use a paid API substitute.