Build Autonomous Content Moderation Agents with Claude Agents SDK
Develop a high-assurance content monitoring agent using Claude Agents SDK designed to detect and flag non-compliant media in real-time. This system utilizes the extended reasoning capabilities of Claude Opus 4.6.6 to evaluate complex visual and textual inputs against strict safety guidelines. The orchestration involves setting up a sandbox environment via Daytona to isolate processing tasks, while Galileo provides continuous observability and automated policy evaluation to ensure the system remains resilient against emerging adversarial patterns. This challenge focuses on building reliable, auditable, and secure agentic workflows for enterprise-grade content safety.
What you are building
The core problem, expected build, and operating context for this challenge.
Orchestrate Claude Opus 4.6.6 agents using Claude Agents SDK to automate content safety monitoring in secure isolated environments.
How work is evaluated
Evaluation of agent accuracy and policy compliance in moderation scenarios.
Challenge datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
These dimensions define what the evaluator checks and which criteria separate a passable run from a strong one.
Compliance Check
Ensure block rate for prohibited content matches baseline.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Decision Accuracy
F1 score of classification • target: 0.95 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
What you should walk away with
Master the Claude Agents SDK lifecycle for managing autonomous tool use and extended thinking processes
Integrate Galileo into the agent workflow to provide real-time policy evaluation and tracing
Build a secure processing sandbox for media analysis using Daytona for runtime isolation
Design user interfaces in Bolt.new to visualize agent decision-making logs and content flagging events
Orchestrate complex task flows where Claude Opus 4.6.6 acts as a primary analyzer and decision-maker
Implement robust error handling for multi-modal input processing within the agent framework
Reference links and supporting material
Sample data for 1 tasks
Challenge setup
Evaluation of agent accuracy and policy compliance in moderation scenarios.
Challenge input
JSON with media metadata
Daytona
Development environments for agents.
Bolt.new
Browser-based AI full-stack app builder.
Galileo
Generative AI eval and observability platform.
Evaluated output
JSON with classification label
- Ensure block rate for prohibited content matches baseline.
- F1 score of classification • target: 0.95 • range: 0-1
- Decision Accuracy target: 0.95
- 1 public reference case
- Python execution harness
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
- Daytona · Required
- Bolt.new · Optional
- Daytona · Required
- Bolt.new · Optional
- Galileo · Optional
- Galileo · Optional
Evaluation contract
- Compliance Check · Weight 1
- Decision Accuracy · Weight 1
Metrics
- Decision Accuracy · Target 0.95
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON with media metadata
Output: JSON with classification label
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.