Back to Ecosystem Pulse
EvaluationExperimental
surfaceloom
by LeonEvo1103
End-to-end testing for agents driving browsers and desktop apps with traceable evidence
TypeScript
Updated Oct 22, 2026
Share:
Summary
Provides an open-source framework to test AI agents driving browser and native desktop apps with explicit tool effects, evidence, lifecycle, and trace contracts. Captures and asserts tool-side effects and evidence chains so tests verify not just outputs but the agent's claimed interactions and UI actions. Built in TypeScript with Playwright-friendly hooks for desktop and web automation, it focuses on observable traces over opaque outputs.
Why It Matters
As agents interact with GUIs and other tools, validating their claimed actions (and the evidence for them) is crucial to trust their behavior. SurfaceLoom makes agent behavior auditable by enforcing lifecycle and trace contracts, which helps build an agent track record and surface reproducible failure modes. Until now many tests asserted only final outputs; this framework treats tool effects and evidence as first-class evaluation artifacts.
When to Use
QA engineers and researchers validating agent-driven UI automation who need reproducible traces of tool effects and evidence for agent-to-agent evaluation.
Real-World Examples
- Validate that an agent's claimed UI actions actually occurred by asserting tool-side effects and evidence
- Reproduce and trace multi-agent interaction failures that involve browser or desktop automation
- Build audit trails for agent behavior to feed reputation systems or continuous agent evaluation pipelines
Works With
playwrighttypescriptwindowsmacos
Topics
accessibilityagent-observabilityagent-testingai-agent-testingai-agentsdesktop-automationmacosplaywrighttest-automationtesting-framework+3 more
Similar Tools
agent-playground
Keywords
agent-to-agent evaluationmulti-agent trustagent track record