Public agent repositories, measured against visible PSF evidence.
PAI scans public GitHub metadata and file paths for signs of production AI discipline: evals, output schemas, observability, deployment gates, human oversight, security policy, and provider resilience. This is evidence coverage, not certification.
Projects are discovered through GitHub repository search, then scanned for visible PSF-aligned evidence in their public file tree. Higher coverage means more evidence was visible to the scanner, not that PAI has certified or endorsed the project.
GitHub public repository search
Repository
Coverage
Grade
Visible evidence
The live benchmark could not retrieve public repositories right now.
Where the repository list comes from
The benchmark uses GitHub's public repository search endpoint and rotates focused queries for AI agent, agentic AI, LLM agent, and MCP server repositories. The run de-duplicates repositories, excludes archived projects and forks when GitHub returns those flags, and sorts the published table by visible PSF evidence coverage.
The benchmark gives maintainers and production AI teams a concrete way to improve visible evidence. A project can publish the missing artifacts, run its own Agent Readiness report, and link to a stable monthly edition when citing broader ecosystem findings.
Use the live table to inspect current public evidence patterns.
Use immutable editions for citations, journalism, and longitudinal comparison.
Use the issue generator to turn a gap into a constructive maintainer task.
Use the evidence pack and control templates to publish the missing artifacts.