ToolTrace Bench v0.3.0 — vendor-neutral, reproducible benchmarking of AI agents.

Measures the whole run: task completion, tool-use correctness, failure recovery,
step efficiency, workspace safety, consistency across repeats, latency
(wall/model/tool) and token/cost data when the adapter reports it. Every run
produces a checksummed .tooltrace bundle that a third party can reproduce.

Runs offline by default. No telemetry, no phone-home, no account.

See CHANGELOG.md for the full list of changes in this release.

Assets

• tooltrace_bench-0.3.0-py3-none-any.whl / tooltrace_bench-0.3.0.tar.gz — the distributions
• sbom.json — CycloneDX SBOM of the declared dependency closure
• SHA256SUMS.txt — checksums for every asset
• frontend.tar.gz — the built dashboard