v0.4.0

Changelog

Features

• 3b1c02d: feat(cli): add judge verdict support to test fixtures (@dpup)
• 49c0cbd: feat(cli): add keep eval command for judge quality measurement (@dpup)
• af42ca9: feat(config): add judge action type and validation (@dpup)
• 5212568: feat(engine): add context.Context to Evaluate and judge action dispatch (@dpup)
• e2c75ac: feat(examples): add judge demo rule file and gateway config (@dpup)
• 15e0f41: feat(examples): add vibe-check judge demo script (@dpup)
• 9bf9443: feat(judge): add Anthropic provider with model shortcuts (@dpup)
• bab2ea0: feat(judge): add Cached field to JudgeAudit and populate from JudgeResult (@dpup)
• b899b52: feat(judge): add Cached field to Verdict type (@dpup)
• 02aa002: feat(judge): add OpenAI provider with model shortcuts (@dpup)
• 90bbcb6: feat(judge): add provider interface and types (@dpup)
• 4bf7b67: feat(judge): implement verdict cache with oldest-first eviction (@dpup)
• f3a7062: feat(judge): propagate Cached field through JudgeResult and adapter (@dpup)
• 51052eb: feat(judge): wire verdict cache in gateway, relay, and demo (@dpup)

Bug Fixes

• fcda727: fix(examples): increase judge timeout to 15s for large context blocks (@dpup)
• 78deb34: fix(examples): use max-turns 3 in judge demo to match gateway demo (@dpup)
• d915853: fix(examples): use self-contained prompt for allowed scenario (@dpup)
• ec42641: fix(judge): bound response body reads and reject unknown decisions (@dpup)
• f7043f6: fix(judge): harden cache and provider implementations (@dpup)
• c5f8f88: fix(judge): use structured outputs instead of free-text JSON parsing (@dpup)

Others

• e66c7a5: docs(examples): add vibe-check judge demo plan (@dpup)
• 5f25f87: docs(judge): add LLM-as-judge implementation plan (@dpup)
• bc36f08: docs(judge): add verdict cache design spec (@dpup)
• 5e0108b: docs(judge): add verdict cache implementation plan (@dpup)
• 9d3d8a7: docs(judge): address plan review findings (@dpup)
• e08aa95: docs(plans): add LLM-as-judge design spec (@dpup)
• 10cb99a: docs(plans): address spec review feedback for LLM-as-judge (@dpup)
• 775e76d: feat(gateway,relay): add judge provider config (@dpup)
• 3757577: test(engine): add comprehensive judge action tests (@dpup)

Full Changelog: v0.3.0...v0.4.0