AI HR Management Toolkit vs Adeu

apply_text_revision

Applies whole-text revised text to a DOCX document by computing a diff and generating tracked changes. Includes a clean-text verification gate to ensure the applied document matches the supplied text. `revised_text` must be the complete CLEAN view of the document: read it with `read_docx` (`clean_view=true`, `page='all'`), edit that text, and send ALL of it back. Never CriticMarkup ({++, {--, {>>) — this tool diffs against the clean view, so markup tokens would land in the document as literal prose. Never one page of a paginated extract — everything absent from the text is applied as a tracked deletion. INTERLOCK: a revision that drops >50% of the characters (>75% for documents under 2000 characters) is refused unless you pass allow_major_deletions=true. If the applied document's clean text does not then match `revised_text`, NOTHING is written to output_path: a diagnostic copy is kept at <name>.unverified.docx and the call fails. output_path defaults to <name>_redlined.docx; an existing _redlined/_processed artifact is revised in place. [Adeu v3.0.6+df9e00c]

diff_docx_files

Compares two DOCX files and returns a compact `@@ Word Patch @@` diff — Adeu's token-level, sub-word patch format — of their text content. Useful for analyzing differences between versions before editing. [Adeu v3.0.6+df9e00c]

finalize_document

Prepares a document for external distribution or e-signature. Note: in this zero-dependency environment, protection_mode='encrypt' is unsupported and falls back to a native read-only lock; export_pdf and password are ignored. [Adeu v3.0.6+df9e00c]

process_document_batch

Applies a batch of edits and review actions to a DOCX. Batches apply SEQUENTIALLY: each change validates against state from prior changes. Valid changes apply when others fail (salvage default): response LEADS with `PARTIAL: applied K of N` listing unapplied changes. Pass partial=false for all-or-nothing. Each item in `changes` needs a `type`: 1. 'modify': search-and-replace. `target_text` must match uniquely (`match_mode`:'strict', default) — add context or set `match_mode`:'first'/'all'. Set `regex`:true for regex matching (groups in `new_text` as $1, $2…). `new_text` supports Markdown: '#'–'######' headings, '**bold**', '_italic_', '\n\n' paragraph split. Omit it (with a comment) to annotate without changing text; empty string deletes. Never write CriticMarkup manually — use `comment`. • EMPTY CELLS: blank cells carry `{#cell:<id>}` anchors — set `target_text` to the anchor and value in `new_text`. Pipes are display separators. 2. 'accept'/'reject': finalize or revert a tracked change by `target_id` (e.g. 'Chg:12'). 3. 'reply': reply to a comment by `target_id` (e.g. 'Com:5') with `text`. 4. 'set_field': fill a form field — `field` is its 'CC:<N>' id, tag or alias, `value` the text; list via `read_docx` `mode`:'fields'. Checkboxes take true/false, dates YYYY-MM-DD, dropdowns a listed option. Dual-writes bound stores. A locked/protected control refuses and names the override permitting it. 5. 'insert_row': add table row — `target_text` anchors on an existing row's text, `cells` holds cell values (left-to-right), `position` is 'above'/'below' (default below). 'delete_row': remove row matching `target_text`. Disk mode only. ID VOLATILITY: 'Chg:N'/'Com:N' ids shift between states — call `read_docx` before accept/reject/reply; never reuse ids from earlier turns. `author_name` sets Track Changes attribution; defaults to 'Adeu AI (TS)' when omitted. [Adeu v3.0.6+df9e00c]

read_docx

Reads a DOCX file. Returns text with inline CriticMarkup for Tracked Changes and Comments: {++inserted++}, {--deleted--}, {==highlighted==}{>>comment<<}. Set clean_view=True for the finalized 'Accepted' text without markup. Modes: - 'full' (default): paginated body content. Use page=N to navigate. - 'outline': heading map only — start here for large docs to plan targeted reads. Defaults to L1-L2 headings; pass outline_max_level=3-6 to see deeper structure. - 'appendix': defined terms, anchors, and cross-reference targets. Consult before editing legal/technical docs to avoid breaking references. - 'changes' (mode='changes'): a ledger of every tracked change and comment (id, type, author, page, snippet) — start here for review work instead of reading pages. Filter with changes_author, page, and changes_offset. - 'fields' (mode='fields'): a ledger of every content control (ordinal, class, alias/tag, location, lock/binding state, current value) — start here to discover fillable fields. Paginate with fields_offset; `page` and `search_query` do not apply. `page`: a positive integer (1-indexed, default 1), a page RANGE like '2-6' (returns up to 8 pages in one call, then names the next range), or 'all'. Pages are synthetic length-based chunks sized for LLM consumption, NOT printed Word pages. In mode='full', page='all' returns the whole body with no page chrome; oversized documents are refused with an outline and a bounded-read recipe unless force=true. With `search_query`, `page` instead restricts matches to that page (default: search all pages). [Adeu v3.0.6+df9e00c]