Build an input sanitizer for the chatbot

[sahilds1]

This ticket relates to #411 (comment)

[sue-hntr]

This is the task to be completed for #437, pulled from #411:

Release 1 — Pronouns + Exact No-Match + Minimal Linter
This step locks in the exact wording we want and adds a tiny safety net to clean up any stragglers.
Ship two small instruction lines (canonical attribution + exact no-match sentence) to set the baseline.
Add a lightweight regex-based UI wording linter on final text only to normalize “your documents/files/sources.”
Possible testing step: Validate with unit tests and a short batch run; enable via feature flag with a one-toggle rollback.

For full information, please see these docs and code in #411:

AI-Slides_How-we-keepAskJJ-OutofTrouble.pdf

AskJJ Rollout Plan - Jan 2026.pdf

ui_linter_rules_v01.py

test_ui_linter_v01.py

Please note that, as a UX designer, this recommended file from ChatGPT is outside of my scope. Chat phrased it as a "nice to have". I have been the only one documenting AskJJ responses and only during pre-release on https://balancertestsite.com/. I have no clue how this test file would work.

#3 Markdown doc for release 1: ui_linter_README.md

ui_linter_README.md

[sue-hntr]

I need to wait until #437 is complete and tested before I continue on with the following 2 releases.
Each release depend on the work from the previous one (The plan for #411 is cumulative)

Release 2 — Self-Awareness + Tone/Format
This step makes AskJJ sound polished and consistent without changing what it says.
Add a single self-identification line so AskJJ clearly references Balancer and its corpus.
Tighten structure (clear headings, concise paragraphs, neutral tone) without expanding the linter unless a recurring issue appears.
Re-run the same tests to confirm no regressions; no flag changes needed.

Release 3 — Edge Templates & Fallbacks
This step aligns the rare paths (errors, no-results, fallbacks) with the new wording and style.
Update those templates to use the approved attribution and the exact no-match sentence.
Add only narrowly scoped linter rules for proven variants discovered in real usage.
Confirm zero forbidden phrases, low linter hit-rate, and stable performance—then consider trimming the linter.

[sue-hntr]

Hi @AkhilRB0204 and @sahilds1
I see that this has been picked up!
Please let me know when this is ready for review -- not by code but by the AskJJ user interface.
Thanks!