[kklimuk]

🤖 docx-cli triage

What was wrong

A first slice of #40: the two bookmark-integrity bugs that don't depend on the field model (#41). On tests/fixtures/multi-column.docx, p103 holds _Ref279752219, which p37 and p93 cite:

• edit --at p103 "Psy. 2013. *Gangnam Style*. Video." dropped the bookmark (17 → 16 pairs).
• edit --at p103 --text "…" kept the bookmarkStart but dropped the bookmarkEnd, which sat inside the rebuilt <w:hyperlink> (17 starts, 16 ends).

Both exit 0 and pass validate, but every REF to the bookmark prints "Error! Bookmark not defined."

Separately, raw insert --xml '<w:p><w:bookmarkStart w:id="100" …/>…<w:bookmarkEnd w:id="100"/></w:p>' landed as start id 100 and end id 0.

What changed

• New src/core/bookmarks/markers.ts. It lifts a paragraph's bookmark markers (direct children, or inside run-bearing wrappers) and restores them: a start goes right after <w:pPr>, an end after the last child. It keeps the original nodes, so every attribute survives.
• Edit.paragraph (single and batch, --markdown/--runs/--text rebuild):
  • It lifts the markers and puts them back on the first new paragraph. That is the same carrier lifted object runs use.
  • Under tracking they go onto the transition paragraph, outside the <w:del>/<w:ins>, so both accept and reject keep the bookmark.
  • On the --text word-diff path, markers that survived stay where they are, and only casualties come back at the paragraph boundary.
  • A cross-paragraph bookmark keeps its half on the edited paragraph.
• remintBookmarkIds (src/core/raw/references.ts): the first time it sees a free id, it records an identity mapping, so the end marker reuses its start's id.
• Tests: tests/cli/edit-bookmarks.test.ts (6 tests) and one raw-gate test. The 4 untracked edit tests and the raw test fail on main.

Verification

• Before: e1 starts=16 ends=16, e2 starts=17 ends=16. After: all 17/17, with w:id="13" paired on p103. This holds for untracked, --track + accept, and --track + reject, and validate passes on all of them.
• Raw insert: start and end are both w:id="100".
• bun run check ✅ bun run test:unit ✅ (2098 tests)
• /simplify (one pass in this context, without the 4-agent fan-out): rewrote lift as collect-then-remove. /code-review xhigh: the forked run reviewed the wrong checkout, not this diff, so I did a manual review instead. That review led to two changes: restore now keys markers by tag+id, so a cloned wrapper can't duplicate a marker, and removal now splices in place.

Decisions for you

• Nothing in this PR. The rest of feat(bookmarks): bookmarks as addressable targets, authored cross-references (REF/PAGEREF/NOTEREF), and internal links #40 is blocked on feat(fields): model complex fields — parse, AST, read tokens, offset widths #41 and on the questions in the issue comment.

Part of #40 (not a full fix; the issue stays open).

---

Generated by Claude Code