draft-refine

# draft-refine Follow the [three-tier output convention](../../core/CONVENTIONS.md#5-skill-output-convention--t1t2t3). Store review logs in `_internal/strategy_reviews/` or `_internal/draft_reviews/`. Use `_internal/versions/v{N}/` for modern snapshots and recognize legacy sibling `_v{N}.md` snapshots when already present. ## Resolve the Document Resolve `$ARGUMENTS` to a canonical strategy or draft and any existing required audience-language companion. - A path containing `/draft/` selects draft mode and canonical `draft.md`. - A path containing `/strategy/` selects strategy mode and canonical `strategy.md`. - A directory defaults to `strategy/strategy.md`. - A Markdown file is used as supplied; recognize a sibling `draft_ko.md` or `strategy_ko.md` only when it exists or the workflow explicitly requires that legacy companion. - Otherwise fuzzy-search with `ls -d <artifact-root>/documents/*$ARGUMENTS* 2>/dev/null`. - one match → resolve its canonical strategy and existing companions - multiple matches → ask the user to choose - no match → report the resolution error Do not generate a language-companion pair by default. Preserve the target artifact's existing or explicitly requested language and follow `<agent-home>/roles/response-policy.md` for user-facing output. ## Prepare Versioning Before dispatching the `research/research-survey` unit, prepare every existing canonical or companion file that the refinement will change through the active route's write prefligh...