Perfect AI Beauty Agent: The A2A Integration Guide for Beauty Brands

Architecture

The flow has two sides:

• Client side — End User → Brand Website/App → Client AI Agent loads SKILL.md → invokes Perfect AI Beauty Agent. Image input (selfie / live camera) is captured via Perfect CameraKit or the device's native camera.
• Perfect Corp side — Perfect AI Agent routes through: LLM (OpenAI or LLM of choice) for intent understanding → RAG for brand knowledge, SKU, and recommendation logic (sourced from Perfect Console) → Services Catalog for execution.

Intents (Available Capabilities)

Each capability is exposed as a named intent. Your host agent registers these as tools. The agent runtime selects and invokes the correct intent based on the user's request.

Request & Response

4.1 Request — Input Fields

4.2 Response — Output Fields

4.3 Example: Chained Flow (Skin Analysis → Recommendation)

# Step 1 — Host agent sends skin analysis request{"intent": "skin_analysis",
  "brand_id": "acme-beauty",
  "prompt": "Help me with my skincare routine.",
  "image": "<base64 selfie from CameraKit>"}# Step 1 — Perfect AI Beauty Agent response{"intent_handled": "skin_analysis",
  "message": "Your skin type is Normal. Top concerns: Moisture (mild), Dark Circles (mild).",
  "image": "<annotated skin report URL>",
  "analysis_id": "ana_abc123",
  "data": {"skin_type": "Normal",
    "concern_scores": { "moisture": 68, "dark_circles": 72 }}}# Step 2 — Host agent chains into recommendation{"intent": "product_recommendation",
  "brand_id": "acme-beauty",
  "analysis_id": "ana_abc123",
  "prompt": "Suggest serums for my top concerns."}# Step 2 — Perfect AI Beauty Agent response{"intent_handled": "product_recommendation",
  "message": "Based on your skin report, here are 2 serums for moisture and dark circles.",
  "data": {"recommendations": [{ "sku_id": "SERUM-VC01", "name": "Vitamin C Serum", "match_score": 0.94, "reason": "Targets dark circles and hyperpigmentation" },
      { "sku_id": "SERUM-HA02", "name": "Hyaluronic Acid Serum", "match_score": 0.89, "reason": "Deep hydration for moisture-deficient skin" }]}}

SKILL.md — Manifest Reference

Drop the skill folder into your agent's skills directory. Your agent reads SKILL.md at startup, registers the intents as callable tools, and selects the correct one automatically based on user intent.

Folder Structure

perfect-ai-beauty-agent/
├── SKILL.md                  # manifest — capabilities, auth, data isolation
├── references/
│   ├── intent-schemas.md     # full input/output field definitions
│   └── brand-onboarding.md   # brand setup checklist
└── assets/
    └── arch-overview.md      # A2A architecture notes

Key Frontmatter Fields

---
name: perfect-ai-beauty-agent
description: >
  Invoke Perfect AI Beauty Agent for skin analysis (CameraKit),
  skincare product recommendations, virtual try-on, or beauty consultation.
  Brand-scoped: all data partitioned by brand_id.
  Trigger: skin analysis, skincare advice, product recommendation,
  virtual try-on, beauty consultation, skin score, CameraKit.
license: Proprietary — contact api-support@perfectcorp.com
metadata:
  author: Perfect Corp RD
  version: 0.1-draft
---

Setup & Onboarding

Error Codes

Data Isolation & Security

• brand_id is required on every request. All SKU catalogues, RAG corpora, recommendation rules, and user history are partitioned by brand_id at the data layer.
• API Keys are brand-scoped. A key issued for brand A cannot access brand B's data — requests return 403.
• CameraKit sessions are ephemeral and token-scoped. No video frames or images are retained server-side after the session closes.
• user_id is an opaque token. No real PII is required or stored. Brands opt in to persistent user history — disabled by default.
• API Keys can be rotated any time from the Perfect Console. Old keys are revoked immediately on rotation.