📝 README repositioning as additive fork 2 commits 2mo ago minor change
jamietso/mike-redline
jamietso/mike-redline surfaces tracked changes and comment bubbles to the LLM
jamietso's fork adds redline-aware document extraction to upstream mike. DOCX files are now parsed by a direct word/document.xml walker (replacing mammoth) that emits {++ins++} / {--del--} markers at w:ins/w:del boundaries and inlines comment text at each w:commentRangeStart anchor. PDF redlines are handled by a Python/PyMuPDF script that classifies text color (red = deletion, blue = insertion, green = move) and outputs the same marker scheme. Both the chat assistant and the tabular reviewer see the annotated string; system prompts in both are updated to define the markers.
The README was later reframed to present the fork as a focused additive contribution rather than a gap-filler relative to closed-source tools, and a demo video was embedded.
What's in it
- Tracked changes preserved for the LLM DOCX and PDF documents reach the model with insertions, deletions, moves, and reviewer comments still visible as inline markers - not silently accepted away.
- Redline-aware prompting The system prompts are tuned to understand and reason about the marked-up format, so the model can discuss what changed and who suggested it.
- Additive positioning relative to upstream Presented as a complement to vanilla Mike rather than a competitor - the accepted-view default still makes sense for most readers; this fork serves a specific need.
Direction
contract-reviewdrafting
Activity
📝 jamietso teaches Mike to read the redlines, not the clean copy 1 commit 2mo ago contract-reviewchat-ui draft
The AI now sees insertions, deletions and comments as edits - so it can tell you what the other side actually changed.
Threads of work (detailed view)
jamietso teaches Mike to read the redlines, not the clean copy
The AI now sees insertions, deletions and comments as edits - so it can tell you what the other side actually changed.
jamietso/mike-redline: feed tracked changes and comment bubbles to the LLM
Upstream mike strips redlines before the LLM sees a document - `w:del` text and comment bubbles are silently dropped. This fork keeps them, inlining `{++ins++}`, `{--del--}`, `{<<moved>>}`, and `{>>by AUTHOR: text<<}` markers so the model can answer "what did counterparty change?" on both DOCX and PDF.