Facial Consultant

Facial Consultant

An analysis skill that can call two APIs. There is no run.py: you (the agent) run the flow here through youcam_core.py, and you also do the golden-ratio interpretation from the references.

When to trigger

When the user wants their skin tone, facial features / face shape, or facial proportions. For skin-condition scores, hand off to skin-analysis-expert.

Ask first (this skill has multiple analyses)

Before running anything, ask the user what they want and state the credit cost of each:

• Skin tone — colors of skin/eyes/brows/lips/hair (skin-tone-analysis)
• Facial attributes & ratios — feature shapes, face shape, golden-ratio proportions (face-attr-analysis)
• Both Get the per-call cost with python scripts/youcam_core.py cost --feature skin-tone-analysis and --feature face-attr-analysis, tell the user, and only run the ones they choose.

Get the latest API spec (do this every run)

Read the doc section(s) for the chosen analyses and open the latest version for the current endpoint / parameters:

• Skin tone: https://docs.perfectcorp.com/reference/ai_skin_tone_analysis.md
• Facial attributes / ratio: https://docs.perfectcorp.com/reference/ai_face_analyzer.md If newer than api-fallback.yaml, pass --version.

Prerequisites

1. A clear, front-facing, single-person selfie. skin-tone and face-attr accept jpg/jpeg only (no png).
2. python scripts/youcam_core.py validate-key (0 credits); if the key is not set, guide the user to set it.

Run (only the chosen analyses)

python scripts/youcam_core.py run --feature skin-tone-analysis --src_file  --param face_angle_strictness_level=high
python scripts/youcam_core.py run --feature face-attr-analysis  --src_file  \
    --param face_angle_strictness_level=high \
    --param features='["eyeShape","eyeSize","eyeDistance","eyebrowShape","cheekbones","faceShape","lipShape","noseWidth","noseLength","eyeColor","lipColor","eyebrowColor","hairColor","horizontalThird","verticalFifth","faceAspectRatio","upperLipToLowerLip"]'

Interpret colors and features with references/guide.md, and do the golden-ratio bucketing yourself (buckets and golden targets are in that file).

Output rules

• For colors, report colors only; do not infer an undertone that was not returned.
• Ratios drift slightly between runs and faceShape may not be returned; label ratios as approximate and skip missing values instead of filling them in.
• Always end with this note: "This is an AI-based facial analysis for reference only. It is not a medical or cosmetic-surgery assessment; results can vary with lighting and pose."