T-YF4K-quality-runner-warnings-not-fail

Status: planning/draft · Impact: medium · Complexity: small

Goal

The plain sdlc quality run (no baseline) reports a hard FAIL for a lint check whose underlying command actually exits 0 — when the only findings are pre-existing biome warnings, biome ci returns success but the quality runner’s finding-extractor classifies the warnings as a FAIL. That forces an implementation sub-agent to stash its changes and re-run on the base tree just to prove the FAIL is pre-existing and unrelated to its diff. This task makes the quality runner distinguish biome warnings (exit 0) from errors so a warning-only run isn’t reported as a hard FAIL — and/or steers the implementation-sub-agent brief to always use the baseline-gated form so pre-existing warnings are subtracted automatically.

From T-HPWU-document-cross-module-dts-emit-condition (git@github.com:sksizer/markdown-contract.git): The plain ‘sdlc quality run’ (no baseline) reports ‘FAIL bunx moon run core:lint’ when the only findings are pre-existing biome warnings, even though ‘biome ci’ itself exits 0 — the finding-extractor classifies warnings as FAIL, forcing the implementation sub-agent to stash and re-run on the base tree to prove the FAIL is pre-existing. Fix: the quality runner should distinguish biome warnings (exit 0) from errors so a warning-only run isn’t reported as a hard FAIL (and/or the implementation-sub-agent brief should always steer to the baseline-gated form).