Files
jobtrackingapp/docs/career-workspace-ux-refactor.md
cesnimda 56fed05d70
CI and Deploy / test (push) Successful in 2m42s
CI and Deploy / deploy (push) Failing after 9s
feat: complete phase 2 UX improvements
2026-07-30 21:35:52 +02:00

12 KiB
Raw Permalink Blame History

Career Workspace + CV Builder UX refactor

2026-07-20. UX/product restructuring against v1.0.0. Frontend-only — no change to CareerProfiles, CvVariants, the CV generation pipeline, extraction APIs, AI services, permissions, or tenant isolation.

This document is the plan. It is delivered staged: Phase 0 (the sidebar bug and the framing copy) is implemented and tested now; Phases 14 are scoped for incremental delivery because they require surgery on the 1376-line CareerProfilePage and touch the production CV/extraction flow, where a single large rewrite would risk the "do not break" list. Each later phase is independently shippable and verifiable.

Current problems

  1. Sidebar double-highlight. On /career/builder/{id}, both "Career Workspace" and "CV Builder" light up. Root cause: AppShell used pathname === to || pathname.startsWith(to + "/") per item, so /career matched every /career/... child. No "most specific wins" rule. (Fixed — Phase 0.)
  2. Too many competing concepts on one page. CareerWorkspacePage is a thin shell around CareerProfilePage (1376 lines), which bundles: a structured-CV editor, extraction-run history, a "CV structure overview", rewrite templates with a PDF carousel (a second, template-driven CV builder), and field-level review metadata. A user cannot tell which artifact is "their CV".
  3. Internal vocabulary leaks to users — "structured CV", "JSON structure", "extraction schema", "structure overview". These are implementation concepts.
  4. Extraction is opaque. Upload runs, the profile changes, but the user never sees what changed and cannot approve it. History is shown as a primary section instead of the result of an import.
  5. Two CV builders. The template-driven rewrite/PDF flow inside the profile page overlaps the real CV Builder (/career/builder), which already owns templates, layout, styling, variants and PDF.

New information architecture

Product rule: Career Profile holds your information; CV Builder creates documents from it.

Surface Owns Does NOT own
Career Profile (/career) Personal info, professional summary, work experience, education, skills, projects, certifications, languages. The facts. Templates, layout, styling, PDF, variants
CV Builder (/career/builder, /career/builder/{id}) Templates, layout, styling, section order/visibility, variants, PDF generation Career facts (it reads the profile)

Career Profile page structure (target):

  • Header — "Career Profile", subtitle "This information powers your CVs, applications, cover letters and AI assistance.", profile completeness %, last updated, quick action → CV Builder.
  • Sections (user-facing labels only): Personal information · Professional summary · Work experience · Education · Skills · Projects · Certifications · Languages.
  • Import CV — current source (filename, date, status) + [Upload new CV]; after extraction a review screen (below). History moves to Settings → Advanced → Import history.

Removed / relocated concepts

Concept Disposition
"Structured CV Editor" Renamed and reframed to Career Profile editor (same fields, user vocabulary)
"CV Structure Overview" Removed from the user surface; if needed for debugging, move under admin/developer tools
"Template-driven CV Builder" (rewrite templates + PDF carousel inside the profile page) Removed — the CV Builder already provides templates, layouts, styling, sections, customization and PDF. One CV Builder only
Extraction run history as a primary section Relocated to Settings → Advanced → Import history
Internal terms ("structured CV", "JSON", "schema") in labels/help text Replaced with plain language

Route ownership (authoritative)

/career                 → Career Profile        (Career Workspace nav item)
/career/builder         → CV Builder            (CV Builder nav item)
/career/builder/{id}    → CV Builder            (child of CV Builder, NOT Career Profile)

Rule: the sidebar item whose to is the longest prefix the current path is at or under wins; all others are inactive. A child route never activates a parent nav item.

Implementation plan

Phase 0 — Sidebar bug + framing (DONE, this change)

  • AppShell.activeNavTo(pathname, tos) — exported pure function; longest-owning to wins. selected now compares against the single computed activeTo across both nav lists.
  • Breadcrumb/title in App.tsx: explicit /career/builder → "CV Builder" ownership before the /career fallback (previously /career/builder showed "Career Workspace").
  • Career Workspace header reframed to the "Career Profile" product framing.
  • Tests: sidebar-active-nav.test.ts — asserts exactly one active item for /career, /career/builder, /career/builder/{id}, and that no item double-highlights.

Phase 1 — terminology + first component split (IN PROGRESS)

Delivered 2026-07-20 (increment 1):

Terminology → user-facing (src/i18n/translations.ts, no structural change):

Internal term (before) User-facing (after)
"Structured CV editor" "Career information"
"CV structure overview" "Profile sections"
"Summary bullets" "Professional summary"
"Core skills" "Skills"
"Analyze sections" "Read sections"
"Original extraction" "Original import"
hardcoded "Master career profile" "Career profile"
Help text de-jargoned; the "Career information" help now says *"The CV Builder uses this information
to create documents."*

Component extracted: src/views/career/ProfileCompleteness.tsx — the completeness meter + missing chips + version-history accordion, pulled out of CareerProfilePage. Display-only, props in, no state or API — the first step of the split.

No API / data / model change. The save path is untouched: api.put("/career/profile", { profile: structuredCv, cvText }). A new test (profile-page.test.tsx → "saving the career profile PUTs … unchanged (Phase 1 refactor invariant)") pins exactly that call so the remaining extraction can't silently change it. Existing profile-page tests were re-pointed to the new labels; all behavioural assertions (save, parse, field values) kept.

Verified: tsc clean, production build clean, 136 frontend tests pass (was 135; +1 invariant test). Sidebar fix from the previous task still passes.

Delivered 2026-07-20 (increment 2 — COMPLETED):

Sections extracted into src/views/career/CareerProfileSections.tsx (one file, one component per export), plus FieldReviewNote + confidenceTone moved there verbatim and shared with the parent:

Component Slice Props
PersonalInformationSection contact value, onChange, getMetadata
ProfessionalSummarySection summary value, onChange, getMetadata
SkillsSection skills value, onChange, getMetadata
InterestsSection interests value, onChange, getMetadata
LanguagesSection languages value, onChange, getMetadata
WorkExperienceSection jobs value, onChange
EducationSection education value, onChange
OtherSectionsSection otherSections value, onChange

Each section is presentational: it receives its slice + onChange(next); the parent still holds structuredCv, loads, saves, and owns every extraction/import action. getMetadata is a callback (getStructuredCvFieldMetadata over the whole profile) so sections stay decoupled from the full shape. No section makes an API call. CareerProfilePage dropped from 1376 → ~1200 lines.

No ProjectsSection / CertificationsSection were created — the editor never had those sections (StructuredCvProfile has no editable projects/certifications UI here; such content lives in "Other sections"). Inventing them would add functionality, which this refactor explicitly avoids. Flagged for a product decision in a later phase.

Duplicate concepts hidden (not deleted) behind an "Advanced CV tools" toggle on the Career Profile page, collapsed by default:

  • CV Structure Overview ("Profile sections" parse block)
  • Template-driven CV Builder (rewrite templates + PDF carousel)

Both stay mounted and fully functional — gated with display: none via showAdvancedCvTools, so the underlying flows (and their tests) are intact and reachable, just out of the default workflow. The real CV Builder at /career/builder is the single CV-generation surface.

Tests (all green): existing profile-page tests re-pointed to reveal advanced tools before touching those controls; added "editing a field in an extracted section updates parent state and flows into save" (proves render → edit → PUT /career/profile { profile, cvText }). The save-invariant test from increment 1 still pins the exact payload.

Verified: tsc clean, production build clean, 137 frontend tests pass. No API, save-payload, extraction, or data-model change.

Future removal plan (later phase, not now): once Phase 2 (Import CV review) and the real CV Builder cover every flow the hidden blocks serve, delete the structure-overview parse block and the template-driven builder from CareerProfilePage entirely, and drop the showAdvancedCvTools toggle. Until then they remain behind the toggle so no tested functionality is lost.

Still open for Phase 1 polish (optional, low priority): a dedicated CareerProfileHeader component and per-section actionable empty-state copy ("No work experience added yet" → the Add button already present). Neither is blocking; both are cosmetic.

Phase 2 — Import CV review screen (DONE 2026-07-30)

Extraction runs now stop at pending_review; they no longer overwrite the career profile. A backend diff and conservative merge preserve curated/unmatched data, stable item IDs and non-empty values. The Career Profile shows additions/updates by category with Apply/Discard actions. Low-confidence changes have stable change IDs, are excluded by default in the API, and require individual checkbox confirmation. Projects, certifications, languages-from-prose, grouped skill prefixes, glued date/title text, common mojibake, and nested earlier/part-time roles are covered by regression tests. Part-time roles are separate experience entries. No schema migration was required; pending profiles already fit CvExtractionRun.

Phase 3 — Remove the second CV builder (DONE 2026-07-30)

The hidden template/rewrite/PDF-carousel block and all of its state, helpers, saved-job fetch, preview dialog, and obsolete tests were deleted from CareerProfilePage. Career Profile now links directly to /career/builder, the single owner of templates, layout, variants, previews and PDF generation. The raw import section parser remains under Advanced CV tools as a recovery path.

Phase 4 — WYSIWYG for long-form fields (DONE 2026-07-30)

Reused the existing dependency-free Markdown toolbar for the professional summary, work and education details, custom/project sections, and cover letters. Native textarea undo/redo remains available and stored payloads remain clean strings/string arrays.

Rich editor (bold, lists, links, undo/redo) for professional summary, work descriptions, achievements, projects, cover letters. Store clean HTML or Markdown only — no editor-specific state — and the renderer consumes the same format. Choose a small dependency already compatible with the stack, or a minimal contentEditable wrapper; decide at Phase 4 to avoid a premature dependency.

Preserved (verified not touched)

CareerProfiles, CvVariants, CV generation, public CV pages, extraction APIs, AI services, permissions, tenant isolation. Phases 14 are frontend-only; any that appears to need a backend change is a signal to re-scope, not to change the model.

Why staged

The removals and the review/WYSIWYG surfaces all require editing the 1376-line CareerProfilePage and the live extraction/generation path. Delivering them as one change would put the v1.0.0 CV pipeline at risk with no incremental verification. Each phase above is small enough to ship and verify on its own.