Build an Accessible Public Service Application Agent with Claude Agents SDK
Citizens with accessibility needs encounter digital barriers when applying for municipal assistance programs. Implement a Claude Agents SDK workflow using computer use and extended thinking capabilities to automatically navigate public forms, identify accessibility gaps, and complete applications with 95% task completion rate.
What you are building
The core problem, expected build, and operating context for this challenge.
Build an accessible public service agent with Claude Agents SDK utilizing extended thinking to assist disabled citizens with public grant forms.
How work is evaluated
Evaluates the agent's ability to complete public application forms while detecting screen reader incompatibilities.
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.
form_completed_test
Verifies target form was successfully submitted without critical error.
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
completion_rate
Percentage of dynamic form scenarios successfully completed. • 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 Claude Agents SDK extended thinking and tool integration patterns
Implement computer use tools to inspect and fill accessible web form elements
Identify WCAG compliance gaps in digital public service applications
Build resilient form-filling agents that handle multi-step citizen applications
Reference links and supporting material
Synthetic DOM accessibility dataset comprising 25 digital government grant application forms.
Challenge setup
Evaluates the agent's ability to complete public application forms while detecting screen reader incompatibilities.
Challenge input
HTML form snippet or DOM accessibility tree representation
Claude Agents SDK
Provides state-of-the-art computer use and extended thinking agent capabilities.
Evaluated output
JSON containing completed form state and flagged accessibility violations
- Verifies target form was successfully submitted without critical error.
- Percentage of dynamic form scenarios successfully completed. • target: 0.95 • range: 0-1
- Benchmark: WebArenaAccessibility
- Completion Rate target: 95%
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
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
- Confident AI · Required
- Chattforge (Chatforge) · Optional
- Haystack · Optional
Evaluation contract
- form_completed_test · Weight 1
- completion_rate · Weight 1
Metrics
- completion_rate · Target 0.95
Benchmarks
- WebArenaAccessibility
- general
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: HTML form snippet or DOM accessibility tree representation
Output: JSON containing completed form state and flagged accessibility violations
Execution harness: Python
Evaluation tools
- Claude Agents SDK · Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.