LLMKube 0.9.19 fixes agent pipeline that silently bypassed its own safety checks

LLMKube version 0.9.19 was released after developers discovered their AI agent pipeline was producing unverified code reviews while silently skipping the safety rails designed to catch errors. Nine fully tested but production-unreachable files had accumulated on the main branch, hidden because the Go linter counted test references as valid usage. A reviewer agent had been running for three days with no system prompt and a missing tool, leaving it unable to properly evaluate the code it was supposed to review. To address this, the team added a second linter pass that excludes test files, ensuring dead code cannot pass CI undetected, and updated all verdict rails to explicitly log when they cannot run rather than silently passing. The release establishes a design principle that silence and success must always be distinguishable in the system's records.