LegalDocument Review

Collating Reviewer Feedback

Use when users say "collate comments", "combine reviewer markups", "compile tracked changes", "make a resolution checklist", or have multiple DOCX drafts, Word comments, redlines, partner/client marku…

LLegalQuants·Legal·Apache-2.0

Library skill — the default version is maintained in GitHub; edits you make live in your own clone.

Use this skillDownload .zip
How does this work?
  • ChatGPT opens a new chat with the skill loaded. If it's too long for a link, it's copied to your clipboard — just paste.
  • Claude works the same way. To install it permanently, download the .zip and upload it under Claude → Settings → Capabilities → Skills (Pro/Team/Enterprise).
  • Copy prompt copies the skill so you can paste it into any assistant, including Grok.

collating-reviewer-feedback

When to Use

Do not use this skill to automatically accept, reject, or merge changes into the master document. The output is a review checklist. The lawyer makes every document edit.

Audience and Work Shape

Audience: drafting and litigation lawyers, trainees, and paralegals who own or support the master document and understand Word comments/track changes.

Work shape: pattern-matched review with bounded extraction. The skill compiles, groups, and classifies reviewer inputs; it does not decide the legal or drafting outcome.

Legal Failure Modes

Access Modes

This skill works in two practical modes:

  1. File mode - use uploaded or accessible DOCX files, reviewer versions, exports, or extracted Word markup.
  2. User-supplied text mode - use pasted comments, exported revision tables, screenshots, email notes, or manually supplied feedback.

If reviewer files or extracted markup are unavailable, prepare an intake checklist and do not claim to have collated Word comments or track changes.

How It Works

1. Establish the master and reviewer set

Identify:

If the master is unclear, stop and ask. A checklist built against the wrong base document is not reliable.

2. Extract review items

For each reviewer version, extract:

Preserve reviewer identity and original proposed wording exactly. Do not collapse two reviewers into one anonymous note unless the user asks for anonymization. If a table, text box, footnote, field, or other complex Word structure cannot be extracted reliably, create a needs_manual_review item rather than omitting it.

3. Group without pretending to merge

Group items by likely paragraph or clause. Use approximate matching only to organize the review. If matching is uncertain, mark the item as location_uncertain.

Within each group, classify:

4. Produce the checklist

Output a resolution checklist with one row per item:

Field Meaning
id Stable item id
source_file reviewer file or external note source
source_item_id Word comment id, revision id, or external note id where available
source_timestamp comment/revision/note date where available
location paragraph, heading, clause, or best available anchor
location_confidence exact, approximate, location_uncertain, or manual_review
reviewer source reviewer or external source
type classification above
proposal exact proposed text or comment
context nearby master text
status open, accepted, rejected, or deferred
notes lawyer's resolution note

Keep the default status open. Do not infer acceptance from reviewer seniority.

Confidence Bands

Use location_confidence as the operative confidence band: exact, approximate, location_uncertain, or manual_review. Anything affecting tables, footnotes, text boxes, numbering, duplicated clauses, or heavily reworked paragraphs should default down to location_uncertain or manual_review.

5. Call out conflicts and high-risk items

Surface separately:

6. Export for lawyer action

Recommended outputs:

The final step is always: the lawyer edits the master document in Word and checks off the resolution list.

Escalation

Stop and route to the responsible lawyer when a conflict turns on legal substance, including privilege waiver, settlement strategy, regulatory exposure, pleaded relief/remedies, limitation, admissions, or client instructions. Do not classify these as ordinary drafting conflicts.

Example

User: "I have partner, client, and associate comments on the same draft. Can you combine them?"

Response pattern:

  1. Confirm the master draft.
  2. Read each reviewer version.
  3. Produce a grouped checklist.
  4. Flag conflicts and uncertain anchors.
  5. Remind the user that no merged document has been produced.

For a compact output pattern, see examples/output.md. For the detailed item taxonomy, conflict semantics, and paragraph-matching model, see references/reviewer-markup-model.md.

Limitations