amal66 puts legal-document handling through its paces

A new testing layer is designed to make the platform less likely to mishandle the documents, citations and workflows lawyers rely on.

draftingworkflow

This is not a new user feature, but it is meaningful product hardening. amal66 has added checks around the places where a legal-AI product can lose trust quickly: a Word document that comes back wrong, a response that loses its citation rules, or an upload handled unpredictably.

Tracked edits in Word files are checked from creation through acceptance or rejection, including limiting edits to the intended text.
Unclear or malformed editing instructions now have tests confirming the product should stop and report the problem rather than guess.
Upload handling, citation and safety guidance, and the built-in workflow catalogue are all checked for consistency.

The team also raised the testing threshold, making it harder for later changes to quietly erode this protection.

So what Legal teams evaluating AI-assisted document work should care because reliability around edits and evidence is as important as the quality of the first draft.

View this fork on GitHub →

Spotted something wrong? Or know the PR text has fresher detail than the writeup above?

Commits in this thread

1 commit from amal66/mike, oldest first. Source extracted verbatim from the harvested git log.

SHA Subject Author Date

7f09e475 test(backend): document-integrity tier - docxTrackedChanges, prompts, documentTypes, systemWorkflows Amalanand Muthukumaran 2026-07-25 ↗ GitHub

SHA	Subject	Author	Date
`7f09e475`	test(backend): document-integrity tier - docxTrackedChanges, prompts, documentTypes, systemWorkflows	Amalanand Muthukumaran	2026-07-25	↗ GitHub
commit body Four new unit suites (46 tests) for the document-integrity tier of the coverage backlog: - docxTrackedChanges: in-memory docx fixtures via JSZip; insert/delete runs, minimal tracked spans, accept/reject round-trips, id numbering, ambiguous/not-found/malformed-input paths (19 tests) - chat/prompts: base prompt invariants, citation contract markers, and CourtListener splice on/off (8 tests) - documentTypes: type predicates, pdf-conversion policy, MIME fallbacks (12 tests) - systemWorkflows: unique builtin ids, well-formed metadata/columns, non-empty skill markdown, assistant-list sync (7 tests) Coverage ratchet raised from 11/10/14/10 to 23/17/23/23 (measured 23.88% stmts / 17.98% branches / 23.06% funcs / 23.79% lines); backlog doc updated to match. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

commit body

Four new unit suites (46 tests) for the document-integrity tier of the
coverage backlog:

- docxTrackedChanges: in-memory docx fixtures via JSZip; insert/delete
  runs, minimal tracked spans, accept/reject round-trips, id numbering,
  ambiguous/not-found/malformed-input paths (19 tests)
- chat/prompts: base prompt invariants, citation contract markers, and
  CourtListener splice on/off (8 tests)
- documentTypes: type predicates, pdf-conversion policy, MIME fallbacks
  (12 tests)
- systemWorkflows: unique builtin ids, well-formed metadata/columns,
  non-empty skill markdown, assistant-list sync (7 tests)

Coverage ratchet raised from 11/10/14/10 to 23/17/23/23 (measured
23.88% stmts / 17.98% branches / 23.06% funcs / 23.79% lines); backlog
doc updated to match.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Capture this thread into my fork

Download a single Markdown prompt that tells Claude how to port every commit above into your working tree — adapting paths and structure to match your repo. Run it via claude -p < capture-thread-1323.md from inside the repo you want the changes in.

⬇ Download capture-thread-1323.md