# Agent Reading Test

> A benchmark website that tests how well AI coding agents can read and
> interpret web content. Each test page targets a specific failure mode
> observed in real agent-documentation interactions.

## Start Here

- [Agent Instructions](https://agentreadingtest.com/start/): Entry point with test instructions and reporting format

## Test Pages

- [Truncation Test](https://agentreadingtest.com/tests/truncation/): Tests content truncation at known character positions (~150K page)
- [Boilerplate Burial](https://agentreadingtest.com/tests/boilerplate-burial/): Tests whether agents find content buried after ~80K of inline CSS
- [SPA Shell](https://agentreadingtest.com/tests/spa-shell/): Tests whether agents see client-side rendered content or just the static shell
- [Tabbed Content](https://agentreadingtest.com/tests/tabbed-content/): Tests whether agents read all language variants in serialized tabs
- [Soft 404](https://agentreadingtest.com/tests/soft-404/): Tests whether agents recognize a 200-status error page
- [Broken Code Fence](https://agentreadingtest.com/tests/broken-code-fence/): Tests markdown parsing with unclosed code fences
- [Content Negotiation](https://agentreadingtest.com/tests/content-negotiation/): Tests whether agents request markdown when available
- [Cross-Host Redirect](https://agentreadingtest.com/tests/cross-host-redirect/): Tests whether agents follow redirects to a different hostname
- [Header Quality](https://agentreadingtest.com/tests/header-quality/): Tests whether agents can identify sections when headers are generic
- [Content Start Position](https://agentreadingtest.com/tests/content-start/): Tests whether agents find content buried after navigation chrome

## Resources

- [Answer Key](https://agentreadingtest.com/answers.json): Ground truth for scoring agent results
- [Agent-Friendly Documentation Spec](https://agentdocsspec.com): The spec this benchmark is based on
- [Source Code](https://github.com/agent-ecosystem/agent-reading-test): GitHub repository