Roadmap

highbar 0.1 does what the rest of this site describes. These are the things it deliberately does not do yet, and might next. There are no dates.

• Tool stubs. Replacing a tool's result in a test, so the agent can be tested against a failure without breaking a real system. Today, a clean state per test comes from session hooks.
• Machine-readable reports. JSON and JUnit output for CI dashboards. Today the gate is the exit code.
• Variations in multi-turn tests.
• SDKs in other languages. The trace contract is the stable interface; an SDK is a convenience over it.
• LLM-graded checks, kept apart. A judge for the quality of an answer, as its own kind of check. The deterministic verdict stays what it is: a spec without a judge never needs a model, and a judge never changes what a deterministic assertion decided.

For graded quality today, see the comparison.

Ideas and requests: GitHub issues.