Повторная полировка корейского текста

★ 7.0 · content

humanize-redo is a Claude Code skill that re-runs the strict Phase B polish step of humanize-korean on the most recent session's output, skipping a full pipeline restart. The `/humanize-redo` trigger locates the latest `_workspace/{run_id}/` via Glob matching on `final.md` or `01_input.txt`, then feeds remaining findings into `korean-style-rewriter` with a `target_filter` derived from the user's instructions. Supported adjustments include targeting a specific error category (e.g. translated phrasing, idioms, emoji), limiting scope to a single paragraph, tuning intensity between S1-only conservative and full S1+S2+S3 aggressive modes, or rolling back individual edits via `content-fidelity-auditor`. Output is versioned as `03_rewrite_v2.md` with the previous `final.md` backed up, followed by parallel Phase C validation and a Phase D diff table with updated grades. The loop is capped at three rounds; unresolved findings trigger a `hold_and_report` recommendation for human review.