fix: UX Core functional audit fixes + OffSec bias cases

[manager]

What this does

Functional audit fixes across UX Core, plus new OffSec bias content, gated to dev preview.

Fixes

• Bias page sync: modal state follows the route prop instead of freezing at mount — back/forward lands on the right bias, no more stuck dimmed modal.
• Use-case switch inside the modal is a shallow, no-scroll hash update (was a full ISR refetch + jump to top); OffSec state writes #offsec instead of mislabeling the URL as #hr.
• Share previews: absolute og/twitter image fallback (scrapers drop relative URLs), fixed double .OGTags nesting on the Strapi path, twitter:url is the canonical URL, share links URL-encoded so hashes and ampersands in titles survive.
• Strapi cache: getStrapiBiases gets a 5-minute TTL instead of a permanent module cache, and never caches an empty/incomplete response — a transient Strapi failure no longer 404s every bias page until restart, and ISR picks up content edits.
• Rating integrity: thanks-state only after the vote lands, double-click guard, response.ok check, failed geo lookup no longer aborts the vote.
• Search: zero-hit state renders as empty instead of silently falling back to the full library while claiming "found 0"; clearing the input cancels the in-flight debounce; getSearchLabels takes the locale, so EN plurals and RU declensions actually render.
• Resilience: mentionedQuestionsIds parse failure degrades to no mentions instead of crashing the bias page.
• State init: initUseUXCoreGlobals had zero callers (every persisted PM/HR choice was dead weight, each reload reset to PM); now idempotent and wired on /uxcore mount, with an explicit URL hash winning over restored state.

OffSec

• 10 new OffSec bias cases and 4 card surfaces (call, document, poster, timeline), content rewritten in plain language with fewer email surfaces.
• Gated to dev preview only: switcher, row, view and #offsec hash handler behind one shared isOffsecEnabled helper; a persisted isOffsecView flag is treated as inactive off-dev, so a prior dev session cannot leak the half-finished layer onto staging or prod.

Housekeeping

• Machine-readable font passport added to CLAUDE.md (commit gate requirement).

Notes for review

• The branch is behind dev; merge or rebase before landing.
• OffSec is intentionally invisible outside dev preview — verifying it on staging requires the dev env flag.

[MaryWylde]

@claude review this PR

[github-actions]

Claude finished @manager's task in 4m 36s —— View job

---

Reviewing PR #181

• Gather context (PR body, diff, changed files)
• Review correctness & React best practices
• Review TypeScript quality
• Check project conventions (AGENTS.md)
• Accessibility & UX
• Security
• Styling / SCSS hygiene
• Post inline comments for specific issues
• Post final summary

Summary

Reviewed against origin/dev (45 files, +1623/-364). The fixes are careful and well-scoped — each one matches a real bug with a clear root cause, and several show good defensive instinct:

• src/pages/uxcore/[slug].tsx: modal-sync effect now depends on currentActiveBias.number instead of running mount-only — correctly fixes the frozen-modal-on-back/forward bug. The use-case hash effect switched to shallow: true, scroll: false, and mentionedQuestionsIds parsing is now wrapped in try/catch via useMemo. All verified correct.
• src/uxcore/api/biases.ts: TTL cache with a completeness check (en/ru length ≥ expected) before overwriting the cache is a solid pattern — a transient Strapi hiccup falls back to the last good cache instead of poisoning it. Confirmed hy is correctly excluded from the completeness check since it's a partial override set.
• src/uxcore/api/personas.ts / UXCPLayout.tsx / SavedPersonas.tsx: per-call getAuthHeaders() fixes the stale-token-at-mount bug; addPersona/updatePersona/deletePersona now throw on !response.ok instead of silently returning undefined, and every call site was updated to .catch() and surface saveFailed (translated in en/ru, hy correctly omitted per the repo's fallback-to-English convention) instead of optimistically flipping to "saved". Traced this through — no call site was missed.
• src/uxcore/hooks/useBiasSearch.ts: isSearchActive distinguishes "no query" from "query with zero hits" — checked every real caller of the shared setSearchResults (Search.tsx ×2) and both pass the flag explicitly; the other setSearchResults call sites in UXCGLayout.tsx/BiasSearch.tsx are unrelated local useState, not this hook, so the default-parameter fallback isn't reachable with a wrong value in practice.
• OngoingLayout.tsx: the isSubmittingRef double-submit guard plus submittedAnswer returning false on a dropped request (verified against getUXCatSubmitTest, which does return null on a non-ok response) correctly prevents advancing the test on a lost answer.
• OffSec gating: isOffsecEnabled now checks NEXT_PUBLIC_ENV === 'dev' (was !== 'prod', i.e. previously live on staging too) and initUseUXCoreGlobals forces isOffsecView false whenever the flag is off — so a persisted dev-session flag can't leak the layer onto staging/prod. Gate is applied consistently at the toggle, init, and hash-handler layers.

Findings (posted inline)

1. Convention — OffsecBiasView.tsx: builds className via raw template-string concatenation instead of the classnames (cn) helper AGENTS.md requires for conditional classes (lines ~234–236, 299, 310–317). Mechanical fix.
2. TypeScript — src/uxcore/api/rating.ts:8: let userData: any = {} — an explicit any for a value with only 4 known fields (country, region, city, ip).
3. Styling (minor) — OngoingLayout.module.scss:257: new .submitError rule uses a fresh hardcoded #fb1717 rather than a design token; the file already has legacy hardcoded hex, so this is low severity but worth aligning to keepsimple-style for new code.
4. Flag, not blocking — src/uxcore/data/biasOffsec/types.ts: adds 4 new bias-card kinds and 10 new OffSec bias entries. This is local static content behind the dev-only gate, not the canonical Strapi UX Core corpus, but it's still bias-schema/content and the repo's "UX Core data is canonical" policy asks for explicit approval on such changes — flagging so that's confirmed rather than waved through.

No security, hydration, or SSR issues found (all localStorage/window access is properly effect-gated behind mounted/typeof window checks). No App Router, Tailwind, or state-library violations. Import ordering and barrel patterns look compliant throughout.

Note: as the PR description says, the branch is behind dev — the commit list includes unrelated library/widget history from merges; the diff above is scoped to git diff origin/dev...HEAD so it only reflects this branch's actual changes. Rebase/merge before landing as planned.

[github-actions]

Convention: this component builds className with raw template-string concatenation instead of the classnames helper (cn), which AGENTS.md requires for conditional classes ("Always use classnames (imported as cn)"). Applies here and at lines 299, 310–313, and 315.

className={cn(styles.timelineItem, { [styles.timelineItemFlagged]: item.flagged })}

Fix this →

[github-actions]

let userData: any = {} reintroduces an explicit any. Since only country, region, city, ip are destructured from it, a small inline type (or a shared TUserGeo type matching /api/user's response) would keep this typed without much overhead.

[github-actions]

New hardcoded color #fb1717 for the submit-error text — this file already carries legacy hardcoded hex values, but per AGENTS.md "new code aligns to the keepsimple-style skill" / "Do not invent colors". Worth pulling from the design tokens (or an existing error/crimson variable, e.g. the $ks-crimson family used in OffsecBiasView.module.scss) instead of a fresh literal.

[github-actions]

Flagging per repo policy rather than blocking: this extends the OffSec bias-card schema with four new kinds (call, document, poster, timeline) and 10 new bias entries under src/uxcore/data/biasOffsec/. This is local static content (not the canonical Strapi UX Core corpus that CLAUDE.md's "UX Core data is canonical" section governs), and it's correctly gated behind isOffsecEnabled (dev-only) — but since it's still bias-schema content, please confirm this had the explicit sign-off the repo convention calls out for bias data/schema changes before merging to dev.