AI & Software Engineering
A coding-agent harness for database-driven web applications. The project focuses on automated implementation, verification, evaluation and security.
The goal of this project is to build a harness that supports an AI coding agent in developing and verifying database-driven web applications.
Feature or development task
Analyzes the task and modifies the code
Provides context, permissions and project rules
Runs tests and project-specific checks
Detects failures and sends feedback to the agent
Database-driven target web application
Domain, target software and first OpenCode test run.
CompletedHarness setup, custom tool, README, model selection and context files.
In ProgressVerification loop, custom tools, sandbox and permission concept.
PlannedEvaluation set, eval runner, measurements and improvements.
PlannedSecurity review, prompt injection tests and countermeasures.
PlannedStudyTask is a small web application designed for students to manage tasks and deadlines.
The coding agent implements software changes based on specifications and project context.
Automated tests and checks provide feedback to the agent, allowing it to detect and fix errors.
The harness is evaluated using predefined tasks, success rates, runtime and resource usage.
The project includes permission rules, sandboxing and tests against prompt injection attacks.
The project is currently in development as part of the AI in Software Development module in WS 2026/27.