What's new

Every improvement to Still Can Work, most recent first.

All notable changes to Still Can Work are documented here.

Format follows Keep a Changelog; versioning follows SemVer.

0.15.1 - 2026-08-24

Changed

• AI cost cut ~60% with two-tier model routing: judgment-heavy work (resume analysis, resume builder, fit check, LinkedIn review) stays on Claude Sonnet 5; ranking, action kits, interview prep, cover letters, photo review, and the career compass now lead with Claude Haiku 4.5 (~3× cheaper, same fallback chain behind it). Every route also gained an output-token cap sized to its schema, bounding worst-case spend per request.

0.15.0 - 2026-08-24

Added

• LinkedIn review refine loop: an "Answer the fixes" panel under the results takes the information only you have (certifications, volunteer work, numbers, target roles) and regenerates the whole review — headline options, About rewrite, skills, and checklist rebuilt around your facts, treated as truth with no further invention. One page, start to finished profile.

0.14.2 - 2026-08-24

Fixed

• LinkedIn review skills advice now respects LinkedIn's 5-skill top-skills limit: the featured list recommends up to 5 (was a fixed "pin three"), additions are capped at 5 with guidance to swap out removals first when the list is full, and the limit is stated on the page.

0.14.1 - 2026-08-24

Changed

• Every AI wait now shows a real-time progress view: a live elapsed clock, a continuously advancing progress bar tuned to each feature's typical duration, rotating stage messages, and honest copy — including a "still working, complex ones take longer" note when a run exceeds double the typical time. Replaces the static (and sometimes misleading) "about 30–60 seconds" text across resume analysis, job matching, fit check, LinkedIn review, resume builder, photo check, action kits, and interview prep.

0.14.0 - 2026-08-24

Added

• Profile photo check (/photo + /api/photo): upload your profile picture for an honest AI review of the photograph — lighting, background, framing, expression, attire, and image quality, each rated with a fix — plus what to keep, ordered top fixes, a phone-only retake guide, and a two-second first impression. The review judges the photo, never the person, and never suggests hiding your age.
• Enhance studio on the same page, entirely in-browser (the photo never leaves the device for enhancement): drag-to-reposition crop with zoom, one-click auto-enhance (measured levels), brightness/contrast/color/warmth sliders, circle preview, and an 800×800 JPEG download ready for LinkedIn. The AI review pre-nudges the sliders for your specific photo.

0.13.0 - 2026-08-24

Added

• Native ESD weekly job-search log at /esd, fully part of the app: week editor with all three activity types (employer contact, WorkSource, other), search, CSV backup/import, and official EMS 10313 PDF export via pdf-lib — same browser storage as before, so existing logs and tracker-bridged applications carry over. Replaces the embedded standalone page.

Changed

• Header navigation redesigned: three primary destinations (Resume, Jobs, Fit check) plus an accessible "More" menu for Get to know me, LinkedIn review, My applications, My workspace, and What's new.

Fixed

• LinkedIn PDF reviews no longer die with a raw JSON parse error: heavy AI routes now allow up to 300 seconds (the review was hitting the 120-second function timeout), and every page reads API responses defensively, turning platform timeout/size errors into friendly messages.

0.12.0 - 2026-08-24

Added

• Application tracker (/applications): "I applied for this job" buttons on job matches and in the application kit log each application to a device-only tracker — grouped by week, with status (applied / heard back / interviewing / offer / closed), notes, links, and manual logging for jobs found elsewhere.
• ESD weekly job-search log integration (Washington State): the full ESD Job-Search Log app (official EMS 10313 PDF export, CSV backup, search) is embedded at /esd/ and gated behind a tracker setting — "I claim unemployment benefits in Washington State". Tracked applications can be added straight into the correct ESD week as employer contacts (weeks end Saturday, 3 activities per week), sharing this browser's storage with the embedded app.

0.11.0 - 2026-08-24

Added

• LinkedIn profile review (/linkedin + /api/linkedin): upload LinkedIn's own "Save to PDF" export or paste profile text for a recruiter's-eye audit — recruiter-ready score, the five-second first impression, ready-to-paste headline options and a full About rewrite, skills to pin/add/remove (the exact terms recruiters filter on), experience-section fixes, an age-signal scan, visibility settings and habits, photo/banner checklist, and an ordered do-this-today list. Tailored by age group, career compass, and resume analysis when present. Nothing stored.

0.10.0 - 2026-08-24

Added

• Feedback button on every page (floating, bottom-right): anonymous by default with an optional mood and reply email, saved to a private Vercel Blob store (stillcanwork-feedback) — nothing about the visitor is recorded beyond what they type. Includes a bot honeypot and rate limiting. Admins can read everything via GET /api/feedback with the FEEDBACK_ADMIN_KEY bearer token or browse the blob store in the Vercel dashboard.

0.9.0 - 2026-08-24

Added

• Device-only workspace persistence: resume review, original resume/text for the builder, age group, and career context now survive closing the tab and can be resumed on the same browser at /workspace. A clear-this-device control removes every saved item.
• "Get to know me" career compass at /me: an AI-guided profile of goals, preferred work shape, availability, priorities, constraints, and the visitor's own story. Its compact context now tailors job matching, Fit Check, application kits, interview prep, and strategy action kits beyond the resume alone.
• Anonymous campaign-link builder at /share. Use a non-personal scw_campaign label to see aggregated visits in Vercel Analytics, alongside anonymous completion events for key product flows. No resume, job-description, profile, account, or identifying data is sent as a custom analytics property.

0.8.0 - 2026-08-23

Added

• Vercel Web Analytics (@vercel/analytics) for privacy-friendly, cookie-free visitor metrics.
• Gitleaks secret scanning in CI (.github/workflows/gitleaks.yml) on every PR and push to main, matching the other ChAImpOrg repos.
• Platform-level WAF rate limit: 20 requests/minute per IP on /api/* (deny beyond), backing up the in-app limiter.

0.7.1 - 2026-08-23

Fixed

• Resume Builder no longer guesses years for relative dates in user notes ("this month", "recently") — the request now carries today's date and the prompt forbids attaching any date not explicitly stated or resolvable from it.

0.7.0 - 2026-08-23

Added

• Resume Builder (/rebuild + /api/rebuild + /api/resume-docx): one click rewrites your resume applying every fix from the review — age-bias repairs, robot-proofing, stronger bullets — plus your own notes (new certificates, corrections), with a strict no-invention rule. Outputs an on-screen preview, a plain-language change log, and a downloadable, editable Word (.docx) file in a clean ATS-safe layout. The original resume now stays in your browser session (never on our servers) so the builder can use it.

0.6.0 - 2026-08-23

Added

• "Beat the bots" ATS layer: Fit Check now extracts the posting's exact screening keywords (title variants, tools, certifications), marks which ones the resume already evidences, and says where to honestly add the rest — with explicit anti-deception guidance (no hidden text, no unearned claims). Resume analysis now displays its robot-proofing fixes (previously computed but not shown).

0.5.0 - 2026-08-23

Added

• Fit Check (/fit + /api/fit): paste or upload a job description (PDF/Word/text) and get an honest verdict against your analyzed resume — fit score, strong/good/stretch/poor verdict, matched requirements with evidence, gaps with severity and mitigations, talking points, resume bullets tailored to the posting's language, and red-flag scan of the posting itself (age-coded language, scam markers). Linked from the header.
• "Let's do this" action kits on every guide's Four Moves (/api/move): one click turns a strategy into a personal step-by-step plan — a next-15-minutes action, ordered steps, copy-paste message templates, and a measurable finish line. Personalized from the resume analysis when available.

0.4.0 - 2026-08-23

Added

• Direct Anthropic API as the primary AI path (@ai-sdk/anthropic, ANTHROPIC_API_KEY): Claude Sonnet 5 first, Claude Haiku 4.5 second, with the AI Gateway chain retained as fallback. Bypasses gateway free-tier restrictions entirely.
• Version number in the footer linking to a new "What's new" page at /changelog, rendered from this changelog.
• Per-IP rate limiting on all AI routes (lib/ratelimit.ts): sliding window of 15 AI requests per IP per 10 minutes plus a per-instance global cap of 120/minute, returning friendly 429s. Instance-local by design; platform WAF remains the escalation path.

0.3.0 - 2026-08-23

Added

• Logo: a sunrise-over-horizon mark ("a new chapter, still rising") in the header, plus a matching SVG favicon (app/icon.svg).

Changed

• Display font switched from Fraunces (serif) to Plus Jakarta Sans — a modern geometric sans; body remains Inter.

0.2.0 - 2026-08-23

Added

• New age group: In my 40s ("Prime years, full throttle") — full game plan with myths vs facts, four strategies (trajectory positioning, parallel conversations, network activation, pivot playbook), resume rules, curated boards, and red flags. Available at /guide/forties and in the landing-page picker.

0.1.1 - 2026-08-23

Fixed

• AI calls now run through a model fallback chain (lib/ai.ts): the Vercel AI Gateway free tier restricts premium models (403) and rate-limits others (429), and some free models ignore output schemas. The real request is attempted against each model in order (Claude Sonnet 5 first, then Haiku, Gemini Flash, DeepSeek, GLM, GPT-OSS, Llama), skipping restricted/throttled/schema-failing models, remembering the last one that worked. SCW_MODEL env var forces a specific model. When paid gateway credits are added, the app automatically uses Claude again.
• Cover letter generation switched from streaming to a single response so it can use the same fallback path.

0.1.0 - 2026-08-23

Added

• Initial release of Still Can Work — an AI career coach for semi-retired, older, and returning workers.
• Age-group selection (50s, 60s, 70+, returning after a break) with a tailored game plan per group: myth-vs-fact, four strategies, resume rules, curated age-friendly job boards, and scam/red-flag warnings.
• Resume analysis (/api/analyze): accepts PDF, Word (.docx), plain text, or pasted text; returns strongest skills with market demand, suggested roles with fit scores, a send-it-today readiness score, an ATS ("robot-proof") score, an age-bias exposure scan with fixes, line-by-line rewrite suggestions, and a 30-second elevator pitch. Powered by Claude via Vercel AI Gateway.
• Live job matching (/api/jobs): searches Remotive and Arbeitnow (free, keyless APIs), dedupes, and AI-ranks every listing against the candidate's actual skills with an honest match score, a one-sentence "why it fits", and flexibility signals. Plus one-click keyword deep links to Indeed, LinkedIn, AARP, FlexJobs, and USAJOBS.
• Application kit per job: a streaming tailored cover letter (/api/cover-letter) and an interview prep kit (/api/interview) including confident answers to age-adjacent questions, talking points, questions to ask, and negotiation tips.
• Privacy by design: no accounts, no database, nothing stored server-side; analysis state lives only in the visitor's browser session.
• Accessible, high-contrast, large-type design (Fraunces + Inter) with keyboard-visible focus and a print-friendly report view.