# AI Key Point Extraction

## Objective

Reduce long content to the decisions, facts, and requests that actually matter.

A short list of the salient points in a long document or thread, each tied back to the passage it came from.

## Before You Begin

This feature is being added to an application that already exists and already
works. Do not scaffold a new project, and do not assume a blank slate.

Inspect the codebase first and establish:

- The existing application structure and where code of this kind already lives.
- The framework and version in use.
- The existing design system — colours, spacing, typography, and component conventions.
- Existing UI components you can reuse instead of writing new ones.
- The existing database structure, if this feature needs to persist anything.
- The existing authentication and authorization system, if this feature is user-scoped.
- Dependencies already installed, so you don't add a library that duplicates one.
- The existing test setup and conventions.

Only start writing code once you understand the above. If the application
already implements part of this feature, extend it rather than replacing it.

## Implementation Instructions

1. Apply this where length is the problem: long threads, meeting notes, call transcripts, lengthy tickets, multi-page documents. Content that already fits on a screen does not need extracting.
2. Classify each point by kind — decision, fact, risk, request, open question — so a reader can scan for the category they came for instead of reading a flat list.
3. Carry a reference to the source passage with every point and let the reader jump to it. A point nobody can verify is a claim, not a summary.
4. Copy figures, dates, names, and identifiers through verbatim and verify each one appears in the source before display. Rounding a number or shifting a date inside a summary is the failure mode that destroys trust in the whole feature.
5. Cache the result against a content hash so reopening a thread does not re-run the model, and only regenerate when the underlying content changes.

## UI and UX Requirements

Match the application's existing design system exactly. Reuse its components,
spacing, and typography. This feature should look like it was always there.

## Responsive Requirements

Works on mobile, tablet, and desktop. Touch targets are large enough to hit on a
phone, and nothing overflows horizontally at 320px.

## Accessibility Requirements

- Fully keyboard navigable.
- Correct semantic elements and ARIA roles.
- Visible focus states.
- Meets WCAG AA contrast.
- Dynamic changes are announced to screen readers.
- Respects prefers-reduced-motion.

## Edge Cases

- Numbers, dates, monetary amounts, and identifiers must survive unchanged. Check each against the source text and drop any point containing a figure that cannot be found there.
- Repetition is not importance. A phrase in every message of a thread is usually boilerplate or a signature, and weighting by frequency surfaces exactly the wrong content.
- Keep what was established apart from what was suggested. Marking a proposal as a decision changes the meaning of the thread, so label facts, decisions, and recommendations distinctly.
- Every point needs a link back to its passage, and in a thread that means the specific message rather than the thread as a whole.
- Thin or purely social content should yield two points or none. Padding a list to a fixed length forces the model to invent significance that is not there.
- Long content must be chunked and the per-chunk points merged, with near-duplicates collapsed so the same decision does not appear three times in different words.
- In a thread with mixed permissions, only include content the viewer is allowed to read. A summary is a distribution channel, and it will leak a restricted message as readily as any other.
- If the model is unavailable or returns a truncated list, show the original content with an unavailable notice. Never present a partial list as the full set of key points.

## Testing

Exercise the feature end to end in the running application. Cover every edge case
above, then run the existing test suite and confirm nothing regressed.

## Acceptance Criteria

- [ ] Every point is typed as a decision, fact, risk, request, or open question.
- [ ] Every point links to the specific source passage or message.
- [ ] Numbers, dates, and identifiers in points match the source exactly.
- [ ] Sparse content produces few or no points rather than padded output.
- [ ] Results are cached against a content hash and regenerate only when the source changes.
- [ ] Points derived from content the viewer cannot access are excluded.
- [ ] A truncated or failed run shows an unavailable state rather than a partial list.
- [ ] The feature matches the existing design system.
- [ ] No existing functionality is broken.

## Adaptation Rules

- Match the existing design system. Do not introduce a new colour palette,
  spacing scale, or component library.
- Reuse existing components and utilities wherever they fit.
- Follow the naming, file layout, and code style already present.
- Do not upgrade, replace, or remove existing dependencies to make this
  feature fit. Adapt the feature to the app, not the app to the feature.
- Do not break existing functionality. If a change is genuinely required in
  existing code, make the smallest one that works and say so.
- If something in these instructions conflicts with how the application is
  built, follow the application and explain the deviation.

## Final Verification

Before you report the work as done:

1. Re-read the acceptance criteria above and check each one against what you
   actually built.
2. Run the application and exercise the feature end to end.
3. Run the existing test suite and confirm you have broken nothing.
4. Check the feature on mobile, tablet, and desktop widths.
5. Check keyboard navigation and focus handling.
6. Summarize what changed: files added, files modified, and anything you
   deliberately did differently because of how this application is built.

If any acceptance criterion is unmet, fix it before reporting completion.