New agent addition to adk-samples/python/agents. Name of the agent is 'guidelines-driven-media-gen agent'

[anushasabbineni1]

This Pull Request introduces a new agent sample, guidelines-driven-media-gen, to the adk-samples repository. The agent acts as an automated image generation and validation system that iteratively creates and evaluates images based on text descriptions while adhering to predefined brand guidelines and policies (using a fictional brand, NeuroVibe AI, as a demonstration).

The system utilizes a sequential multi-agent pipeline to ensure outputs satisfy compliance thresholds before completion.

✨ Key Features
Multi-Agent Orchestration: Coordinates specialized sub-agents (Prompt Gen, Image Gen, Scoring, and Checker) to achieve complex workflows.
Contextual Prompt Augmentation: Dynamically enriches user intent by consulting a local brand asset bank (TF-IDF retrieval) before generation.
Automated Policy Evaluation: Scores generated visual outputs against quantitative compliance requirements using advanced analysis benchmarks.
Self-Correction Feedback Loop: Automatically re-tries generation by passing failing attribute feedback back into the prompt loop for iterative refinement.
Enterprise Integration Ready: Designed with modularity to easily upgrade retrieval mechanisms (e.g., Vector Search) and connect with Cloud Digital Asset Management (DAM) systems.

📂 Changes Included
The implementation is located under python/agents/guidelines-driven-media-gen/ and includes:

Core Agent & Sub-Agents
agent.py: Main workflow orchestration.
checker_agent.py: Loop control and evaluation thresholds.
sub_agents/image/: Image generation handling (saving to GCS/artifacts).
sub_agents/prompt/: Prompt optimization and asset search.
sub_agents/scoring/: Visual analysis and policy scoring rules.
Tools
fetch_existing_assets.py: Lexical search for brand assets.
image_generation_tool.py: Integration with Image Generation models.
loop_condition_tool.py: State management for self-correction loops.

Configuration & Data
config.py, prompt.py, policy.json: Declarative templates for behavior.
data/: Brand guidelines PDFs, asset images, and metadata for NeuroVibe AI.

Deployment & Verification
deployment/deploy.py: Script to package (wheel build) and deploy securely to Vertex AI Agent Engine (Reasoning Engine).
eval/test_eval.py: Automated pytests leveraging AgentEvaluator for verifying tools local behavior.

✅ How to Verify
Local Testing
You can interact with the agent locally using the ADK CLI or Web interface:

bash

From workspace root

cd python/agents/guidelines-driven-media-gen

Sync dependencies

uv sync --dev

Run via CLI

adk run guidelines_driven_media_gen

Run via Web UI

adk web

Automated Evaluations:
Run the local test suite (does not require Vertex AI Endpoint calls):

bash
uv run pytest eval

Deployed Testing:
Package and deploy to Vertex AI Reasoning Engine:

bash

Build wheel

uv build --wheel --out-dir deployment

Run deployment script (configured via deploy_config.json)

uv run python deployment/deploy.py

Test deployment

uv run python deployment/test_deployment.py