Agent QC

Agent QC defines how Agent projects are classified, tested, evidenced, judged, and reported.

Use it for runtime CLIs, SDKs, tool gateways, channel agents, UI/TUI/desktop/WebUI clients, browser automation, skills/plugins, schedulers, release packages, and eval suites.

Start here

• What is Agent QC?
• Specification
• Quickstart
• Best practices
• Test techniques and compositions
• Project classification
• Gate matrix
• Interaction surface testing
• Evidence contract
• Flow and taxonomy
• Star project testing systems

What changed in this guidance

Agent QC is a standard protocol, not a single-product checklist. The current guidance draws from Agent UI, Agent Knowledge, Codex, Claude Code local snapshot, OpenClaw, Hermes Agent, Playwright, Vitest, pytest, and Agent Skills documentation style. The latest authoring guidance adds explicit snapshot, smoke, black-box, white-box, gray-box, runtime, UI, skills, and composition recipes.