Independent product / Developer tooling
AgentSpec
A testing framework for non-deterministic AI-agent behavior.
My contributionIndependent product and engineering work on evaluation-oriented workflows.
The problem
Catch behavior changes in agents whose output is not deterministic.
Selected work
What I worked on.
- Test definitions and assertions designed for AI outputs.
- Behavior-diff and machine-readable reports.
- A developer workflow that can exercise HTTP-accessible agents.
A testing tool, not a guarantee that an agent is safe or universally reliable.
Context & attribution
The work, in context.
Independent project. The role described here is my contribution, not a claim of sole authorship of every part of the product.
Technology and visual scale
TypeScriptHTML
Measured: 1,450 code lines / 14 files. Code-volume proxy, not value.
Surface bands: language mix in the local code snapshot, including tools and tests.