Create test data
Turn approved knowledge into questions and expected answers.
Dataset guideWhat problem does LLaDAR solve?
Trying one question in a chat window is hard to repeat or review. LLaDAR connects the source material, test question, Agent response, evaluation result, and report, so you can see what an existing Agent actually did.
One evaluation, four stages
Run the target Agent
Find a public way to use the Agent, then save its responses.
Run-agent guideEvaluate answers
Compare each response with what was expected, then calculate the results.
Evaluation guideReport findings
Turn the saved results into a readable report with limitations and details.
Reporting guideUse your own method when needed
After the provider and target Agent are configured, LLaDAR can run the whole workflow automatically. A local directory containing SKILL.md can tailor how questions are generated, answers are judged, or results are summarized.
Before you start: the Agent needs a runnable public workflow, and model-assisted stages need configured provider credentials.
Choose a learning path
New to LLaDAR
Run the included support demo Agent from start to finish.
Get startedEvaluate your project
Use the run-agent guide for a project or controlled test service you already have.
Test an existing AgentChange the method
Use local Skills to choose records, judge answers, or format a report.
Learn about SkillsWhat is saved?
The workflow saves the generated dataset, responses, execution records, evaluation result, and Markdown report. A successful replay shows that LLaDAR reached the selected Agent entry point; it does not by itself prove that every answer is correct.