Vally 0.3.0

Released on 2026-04-29.

Add --threshold option to vally eval for overriding eval scoring thresholds during a command invocation.
Add optional name field to grader config for human-readable display names in reports.
Add vally eval --keep-executor-session-logs to preserve native executor events.jsonl files under --output-dir for session visualizers such as agentviz.
Remote executor workspace sync contract — finalizeWorkspace() hook on Executor interface enables remote executors to materialize workspace files locally before grading. workspaceStatus on Trajectory enables fail-closed safety for file-based graders. requiresWorkspace metadata on GraderBehavior enables dynamic workspace-dependency detection.
Add --known-domains option to vally lint for URL reference scanning (#157)
Add --allowed-external-deps option to vally lint for structural dependency scanning (#157)
Multi-model eval support (#140)
Grader plugin system — ship graders in separate packages (#141)
Grader improvements — regex matching, cross-platform normalization, better diagnostics (#137)
Eval-authoring meta-eval and document directory support in environment.files (#135)
Workspace caching for multi-trial runs (#132)
JUnit XML reporter for eval pipeline (#128)
Export command and exporter plugin system (#113)

Inject environment.skills into executor and honor .vally.yaml paths.skills (#139, #138)
Flatten multi-trial JUnit XML for CI parser compatibility (#131)
Add actions:read permission for dorny/test-reporter (#130)
Fix bug when path to node contains spaces (#127)