AutoCVApply

Upload once. Apply everywhere.
Stop retyping your life story into every job form.

Website · Chrome Web Store · Firefox Add-ons · GitHub · How it works · Dashboard

Sign up free at autocvapply.com - upload your CV, connect the extension, fill forms in minutes.
_{Job board Auto Apply: LinkedIn, Indeed, Totaljobs, Glassdoor, and Reed - full end-to-end apply from the extension sidebar. On other ATS forms, you review every field and click Submit yourself.}

See it in action

Demo video coming soon. Until then, see How it works on autocvapply.com or sign up and try Draft All on a real Greenhouse or Ashby form.

What is AutoCVApply?

Job applications are a copy-paste endurance test. Workday wants your address. Greenhouse wants it again. Ashby wants a cover letter you've already written three times this week. Every ATS renders the same questions differently - custom widgets, shadow DOM, multi-step wizards, iframe embeds.

AutoCVApply reads your CV once, builds a structured profile, and stamps it onto application forms through a battle-tested Chrome extension - so you spend time on roles that matter, not on retyping your phone number for the forty-seventh time.

Without vs with AutoCVApply

Without AutoCVApply	With AutoCVApply
Retype contact details, education, and skills on every ATS	One structured profile - filled from your uploaded CV
Blank stare at "Why do you want this role?" textareas	Draft All streams AI answers for free-text questions
Skip cover letters because they're tedious	One-click cover letter tailored to the job description
Same generic CV for every posting	Tailored resume draft matched to the role
No idea how your CV reads against the JD	ATS score with keyword and formatting feedback
Pray comboboxes and wizards don't break mid-form	6,229-scenario test corpus - Greenhouse, Ashby, Workday, UK job boards, and more
Obvious bot-like paste fills that trip ATS checks	Anti-bot detection - character-by-character typing, natural pauses, human-like navigation on job board Auto Apply
Click through Easy Apply one job at a time	Job board Auto Apply - LinkedIn, Indeed, Totaljobs, Glassdoor, and Reed from the extension sidebar
Submit applications blindly	You stay in control on ATS forms - we fill fields; you review and submit

Job board Auto Apply (LinkedIn, Indeed, Totaljobs, Glassdoor, Reed) runs end-to-end: search filtered jobs, open each posting, fill every step, and submit. More boards are on the way.

Who it's for

Job seekers applying across multiple ATS platforms in a week
Career changers who need tailored answers without rewriting everything
Privacy-conscious applicants who want a clear account, revocable API tokens, and a published privacy policy
Developers and contributors who care about verified form-fill quality, not just demo GIFs

Why choose AutoCVApply?

Save time - autofill plus AI drafting for the questions that actually slow you down
Stay honest - answers draw from your profile and preferences, not invented credentials
Sound human, not generic - Draft All answers are scored on hundreds of scenarios to catch AI tropes, em dashes, and filler while staying grounded in your CV
Trust the engineering - four-layer fill verification, Playwright smoke tests, and 437 PHPUnit methods
Avoid bot flags - human-like typing, pauses between fields, and natural navigation on job board Auto Apply runs
British Postbox UI - Royal Mail red, navy, warm paper tones. Feels like sending a letter, not filling a spreadsheet

Quick install

Recommended: create a free account at autocvapply.com, upload your CV, then install from the Chrome Web Store or Firefox Add-ons (or download a zip from your dashboard for Edge, Brave, or temporary Firefox sideload).

Production (autocvapply.com)

Sign up and upload your CV
Install the extension from the Chrome Web Store or Firefox Add-ons - or open Dashboard → Extension for zip installs
For zip installs: download for your browser and sideload using the on-screen instructions
Copy your connection JSON (token + api_base) into the extension sidebar

Browser	Install method	Store listing
Chrome	Chrome Web Store → Add to Chrome	Live
Edge / Brave	Store listing or dashboard zip → Load unpacked	Chrome Web Store
Firefox	Firefox Add-ons → Add to Firefox	Live

From source (developers)

Clone this repo, run composer run setup, build the extension with npm run build:extension, then load extension/dist/ unpacked in Chrome. See Getting started (local dev) for full setup.

How it works

flowchart LR
    A["📄 Upload CV<br/>PDF or Word"] --> B["🤖 AI extraction<br/>NanoGPT"]
    B --> C["✏️ Review profile<br/>Dashboard"]
    C --> D["🔌 Connect extension<br/>API token"]
    D --> E["⚡ Autofill + Draft All<br/>Any ATS · any site"]
    D --> L["🤖 Auto Apply tab<br/>LinkedIn · Indeed · Totaljobs · Glassdoor · Reed"]
    E --> F["📝 Cover letters<br/>Tailored resumes · ATS score"]
    L --> F

Step	What happens
1. Post your CV	Drop a PDF or DOCX. Tesseract OCR + NanoGPT extract name, contact, skills, experience, and education into a structured profile.
2. Check the details	Tweak anything we missed - summary, visa status, salary expectations, application preferences.
3. Connect the extension	Install the Chrome or Firefox extension, paste your connection JSON (`token` + `api_base`) from the dashboard.
4. Fill and draft	Autofill fields on any job form. Draft All streams AI-written answers for free-text questions, cover letters, and tailored resumes. You submit when ready.
5. Auto Apply (job boards)	Open the extension sidebar Auto Apply tab, pick LinkedIn, Indeed, Totaljobs, Glassdoor, or Reed, and let the extension search, open each job, fill every step, and submit.

Job board Auto Apply

Full end-to-end applications on LinkedIn Easy Apply, Indeed Apply, Totaljobs Quick Apply, Glassdoor Easy Apply, and Reed Easy Apply - not just field fill:

Step	What happens
Search	Extension runs job search with Easy Apply filters from the sidebar Auto Apply tab
Open	Each matching job opens in a tab; listings without one-click apply are skipped
Fill	Contact info, screening questions, resume steps, and multi-step wizards are filled from your profile
Submit	Continue, review, and submit buttons advance and complete the application

On Greenhouse, Ashby, Workday, and other ATS platforms, autofill and Draft All still work as before - you review and click Submit yourself. More full Auto Apply job boards are on the way.

Features

Module	Feature	What it does
CV parsing	PDF & Word upload	Structured profile extraction with Tesseract OCR + local `pdftoppm` preprocessing
CV parsing	Editable profile	Skills, experience, education, summary, application preferences - you control the source of truth
Autofill	One-click fill	Profile data stamped onto native inputs, comboboxes, radios, checkboxes, multi-step wizards
Autofill	Anti-bot detection	Character-by-character typing on short fields, pauses between applies, hover-and-click navigation on job board Auto Apply
Autofill	Shadow DOM & iframes	Content scripts traverse embedded ATS widgets other extensions miss
Auto Apply	LinkedIn Easy Apply E2E	Sidebar Auto Apply tab: search, open jobs, fill steps, submit applications
Auto Apply	Indeed Apply E2E	Indeed Apply search, contact/resume/questions/review steps, submit via Auto Apply tab
Auto Apply	Totaljobs Quick Apply E2E	Genesis Quick Apply search, fill, and submit from the sidebar
Auto Apply	Glassdoor Easy Apply E2E	Host-page search plus Indeed Apply iframe steps where required
Auto Apply	Reed Easy Apply E2E	Reed search, job detail, modal review, and submit
Application Assistant	Field inventory	AI maps the page's questions to fillable refs before drafting
Application Assistant	Job context	Extracts title, company, and description from the posting
Application Assistant	Draft All	Streams batch answers for unanswered fields (NDJSON)
Application Assistant	Draft field	Single-field AI answer on demand
Application Assistant	Answer quality	Rubric scoring on AI drafts - human tone, profile grounding, banned AI phrases, no em dashes
Documents	Cover letter	Styled, job-specific PDF from your profile + posting (extension Auto Apply + Cover tab)
Documents	Tailored resume	Role-matched resume draft
Documents	ATS score	Keyword and compatibility feedback against the job description
Dashboard	Cover letter settings	Design + font preferences (including Random) with live preview
Dashboard	Usage & billing	Monthly autofill allowance, extension connection, GoCardless subscriptions (UK)
Analytics	Public aggregate stats	Daily totals across all users - no personal application history exposed

Cover letters

AutoCVApply generates styled cover letters on the fly in the extension during Auto Apply and from the Cover tab. On the dashboard, open Cover letter settings to pick a design and font (or Random for either), preview the result live, and use those preferences whenever a letter is attached.

Human-sounding AI drafts

Draft All and Quick Answer use AI, but we score generated answers extensively before shipping changes - not just on a handful of demo forms.

What we check	Why it matters
Grounding	Real employers, roles, and skills from your profile - not invented credentials
Human tone	Banned AI phrases and overused words ("leverage", "proven track record", "I am thrilled to apply")
Formatting	No em dashes or markdown - plain text ready to paste into employer forms
Honesty	Gaps acknowledged instead of fabricated experience

Scoring runs across 190 profile-mapping scenarios, 124 answer-quality scenarios, 150 real ATS form E2E scoring fixtures, and Assist sidebar test cases. Developer audit commands: answer-quality:audit, assist-answer-quality:audit, form-e2e:score (see scripts/extension-benchmark/README.md).

Anti-bot detection

Job sites increasingly flag instant paste fills and robotic click patterns. AutoCVApply fills like a person where it matters:

Behaviour	What it does
Character-by-character typing	Short text fields (names, emails, years, locations) type one keystroke at a time with realistic delays
Paced batch apply	Draft All fills one field at a time with scroll, focus, and pause between each
Human-like job board navigation	Smooth scroll, hover-before-click, and variable delays when opening apply flows and stepping through multi-page forms on LinkedIn, Indeed, Totaljobs, Glassdoor, and Reed
Smart fallbacks	Long free-text answers still paste instantly so multi-paragraph drafts do not take minutes

Supported platforms

Autofill works on most major ATS and employer career sites. Auto Apply runs end-to-end from the extension sidebar on supported job boards (search, fill every step, submit). More boards across the UK, Ireland, US, Canada, Australia, and New Zealand are on the way.

Autofill on ATS and career sites

Workday	Greenhouse	Lever	Ashby
SmartRecruiters	Teamtailor	Oracle	BambooHR
Workable	iCIMS	Trakstar	WordPress

_{Plus many more employer career sites and ATS variants.}

Auto Apply - supported today

LinkedIn	Indeed	Totaljobs	Glassdoor
SimplyHired	Reed	CV-Library

Full end-to-end apply: search filtered jobs, open each posting, fill every step, and submit from the extension sidebar Auto Apply tab.

Auto Apply - coming soon

Adzuna	APS Jobs	CareerBuilder	CareerOne	Civil Service Jobs
Dice	Eluta	EthicalJobs	Find a Job	FlexJobs
Government Jobs	Guardian Jobs	Idealist	IrishJobs.ie	Job Bank
Jobs.ac.uk	Jobs Go Public	JobSearch	Jobs.ie	JobsIreland
Jobserve	JobsDB	JobStreet	Jobillico	Jora
Ladders	LG Jobs	LinkUp	Monster	NHS Jobs
NZ Government Jobs	Publicjobs.ie	SEEK	Snagajob	Trade Me Jobs
USAJobs	Workopolis	ZipRecruiter

_{+ more boards across the Anglosphere.}

Note: On ATS forms (Greenhouse, Ashby, Workday, etc.) autofill and Draft All fill fields for you - you review and click Submit yourself. Job board Auto Apply completes submissions end-to-end on supported boards above.

Regenerate this block after changing platform lists in resources/js/lib/site.ts:

npm run readme:platforms

This rewrites the block between  and  in this file.

Fill verification depth (ATS)

Platform	Coverage	Notes
Ashby	Curated + smoke + widget checks	Yes/no and checkbox widget scenarios
Greenhouse	Curated + smoke	Scraped real boards in corpus
Lever	Curated + smoke	Multi-step apply flows
Workday	Curated + smoke	Wizard-style applications
SmartRecruiters	Curated + smoke	Long multi-section forms
Teamtailor	Curated + smoke	Nordic/EU hiring stacks
BambooHR	Curated tier	HR suite apply pages
Trakstar	Curated tier	Performance/hiring forms
WordPress / WPForms	Curated + smoke	Generic employer sites
Any site	Extension runs on `<all_urls>`	Deepest test coverage on ATS platforms above - not where we stop

Postbox design

British utilitarian UI - Royal Mail red, navy, warm paper tones. Built to feel like sending a letter, not filling in a spreadsheet.

Pricing

Plans are based on extension autofill allowance. CV upload and profile editing are free on every plan.

Plan	Price	Autofills / month
Free	£0	250
Starter	£7/mo	2,500
Pro	£17/mo	15,000

Each successfully filled form input uses one autofill. Allowances reset on the 1st of each month.

Security & privacy

Your profile, your account - CV and structured profile data live on autocvapply.com under WorkOS authentication. You can delete your account from settings.
Extension fills locally - the browser extension fetches your profile via a revocable Sanctum token and writes values into the page DOM. It does not send completed submissions back to us.
No data selling - we do not sell personal data. See the full privacy policy.
AI processing - CV parsing and drafting send text to our NanoGPT provider as needed to extract fields or generate answers. Job context from the page may be included in draft requests.
Source available - PolyForm Noncommercial-licensed core. Inspect the extension, backend, and 6,229-scenario test corpus on GitHub. Free for personal and non-commercial use; commercial use requires permission.
You submit on ATS forms - autofill and Draft All never auto-click Submit on Greenhouse, Ashby, Workday, and similar sites. Job board Auto Apply (LinkedIn, Indeed, Totaljobs, Glassdoor, Reed) completes submissions end-to-end from the sidebar; more boards are on the way.

Testing overview

AutoCVApply is verified by two complementary test pyramids - not a handful of smoke tests:

Pyramid	Scope	Fast tier (CI on every push)
ATS form corpus	6,229 extraction scenarios (3,942 vetted), 124 curated fill-verify cases, 13 platform smoke fixtures	Curated JSDOM (70) + Playwright smoke
Job board Auto Apply	212 LinkedIn captures, 44 Indeed step fixtures, 1,700 synthetic job-board scenarios (Totaljobs, Glassdoor, Reed, SimplyHired, CV-Library), unit + offline corpus tests	Unit tests + offline corpus

Both pyramids enforce 100% pass rates on critical tiers before merge. Full tier breakdowns, commands, and fixture layout are in Form corpus quality engineering and Job board Auto Apply testing below.

Job search tips

Practical advice for high-volume applications (no magic numbers - your mileage varies):

Apply while the posting is fresh. Early applicants often face less competition; automation helps you move faster without cutting corners on quality.
Keep your master CV ATS-simple. Single column, standard headings, no graphics in the parse path. Use AutoCVApply's ATS score against the job description before you submit.
Tailor the narrative, not just keywords. Draft All and cover letters work best when your profile summary and application preferences reflect what you actually want.
Review every field. We fill aggressively; you confirm accuracy - especially salary, visa, and eligibility questions.
Track applications yourself for now. Use your own spreadsheet or notes; a personal application tracker in the dashboard is on the roadmap.

Links


Website	autocvapply.com
Dashboard	autocvapply.com/dashboard
How it works	autocvapply.com/how-to
Analytics	autocvapply.com/analytics (JSON)
Blog	autocvapply.com/blog
Privacy	autocvapply.com/privacy
Contact	autocvapply.com/contact
GitHub	github.com/tmwclaxton/autoapplycv
Discord	discord.gg/DqqqTv3Spt - extension help & community
Chrome Web Store	chromewebstore.google.com/.../autocvapply
Firefox Add-ons	addons.mozilla.org/.../autocvapply
Form corpus docs	`scripts/form-corpus/README.md`
Extension bridge (MCP)	`scripts/extension-bridge/README.md`
Job board automation playbook	`docs/platform-automation-playbook.md`

Architecture

flowchart TB
    subgraph Browser["Chrome Extension (MV3)"]
        CS["Content scripts<br/>form-heuristics · field-inventory"]
        SP["Side panel<br/>assist · documents"]
        BG["Service worker<br/>connection · file transfer"]
    end

    subgraph Laravel["Laravel 13 Backend"]
        WEB["Inertia + Vue dashboard"]
        API["Sanctum API<br/>profile · assist · autofill"]
        AI["NanoGPT<br/>CV parse · draft · score"]
        DB[(PostgreSQL)]
    end

    CS <-->|"field inventory<br/>apply answers"| SP
    SP -->|"Bearer token"| API
    BG -->|"Bearer token"| API
    WEB -->|"WorkOS SSO"| WEB
    API --> AI
    API --> DB

Component	Role
`extension/src/content/`	DOM heuristics, field inventory, iframe traversal, portal bar
`extension/src/sidepanel/`	Connection setup, Draft All UI, document uploads
`app/Services/ApplicationAssistantService.php`	Inventory, job context, streaming draft-all, cover letters
`app/Services/CvParserService.php`	PDF/Word ingestion, OCR, NanoGPT structured extraction
`scripts/form-corpus/`	Synthetic corpus generation, fill verification pyramid, E2E harness

Tech stack

Layer	Technology
Backend	Laravel 13, PHP 8.5
Frontend	Inertia v3, Vue 3, Tailwind CSS v4
Auth	WorkOS (web), Laravel Sanctum (extension API)
AI	NanoGPT (`gpt-4.1-mini`) - CV extraction, drafting, ATS scoring
OCR	Tesseract + poppler (`pdftoppm`) locally; NanoGPT vision as fallback
Payments	GoCardless (UK Direct Debit subscriptions)
Extension	Chrome MV3 - content scripts, side panel, service worker
Fill verification	JSDOM, Playwright, pixelmatch, Tesseract.js
Routing	Laravel Wayfinder (typed TS route helpers)

Project structure

autocvapply/
├── app/
│   ├── Http/Controllers/       # Web + API + billing + webhooks
│   ├── Models/                 # User, CvProfile, CvUpload
│   └── Services/               # CV parser, Application Assistant, NanoGPT
├── extension/
│   ├── src/content/            # form-heuristics.js, field-inventory.js
│   ├── src/sidepanel/          # Connection, Draft All, documents
│   └── dist/                   # Built extension (load unpacked)
├── scripts/form-corpus/        # Corpus generation + fill verification pyramid
├── scripts/extension-bridge/   # Localhost bridge + MCP server for agent-driven Chrome
├── resources/js/
│   ├── pages/                  # Inertia pages (Welcome, Dashboard, Billing…)
│   └── components/postbox/     # Shared Postbox UI components
├── tests/
│   ├── Unit/Extension/         # 11 extension test suites (fill, E2E, extraction)
│   └── fixtures/
│       ├── form-extraction/    # 6,229-scenario corpus (html, expected, manifest)
│       └── extension-e2e/      # E2E mocks, scenarios, reports
└── config/subscriptions.php    # Plan tiers and token limits

Getting started (local dev)

Prerequisites

PHP 8.5+, Composer
Node.js 20+, npm
PostgreSQL (or SQLite for quick local dev)
Docker Sail optional

Install

git clone https://github.com/tmwclaxton/autoapplycv.git
cd autoapplycv

cp .env.example .env
composer install
npm install

php artisan key:generate
php artisan migrate
npm run build

Or use the one-shot setup:

composer run setup

Environment

Copy .env.example to .env and configure:

APP_URL=http://localhost

# WorkOS - required for login
WORKOS_CLIENT_ID=
WORKOS_API_KEY=
WORKOS_REDIRECT_URL="${APP_URL}/authenticate"

# NanoGPT - required for CV parsing
NANOGPT_API_KEY=

# GoCardless - optional, for paid subscriptions
GOCARDLESS_ACCESS_TOKEN=
GOCARDLESS_WEBHOOK_SECRET=

Run locally

composer run dev

Starts the Laravel server, queue worker, log tail, and Vite dev server together.

With Docker Sail:

./vendor/bin/sail up -d
./vendor/bin/sail npm run dev

Visit http://localhost.

Build the browser extension

npm run build:extension

The build uses APP_URL from .env only to exclude your local dashboard from content-script injection. The extension API endpoint comes from the dashboard connection JSON (token + api_base).

Then in Chrome:

Open chrome://extensions
Enable Developer mode
Click Load unpacked
Select the extension/dist/ folder

Generate a connection from the dashboard (Copy includes token + api_base) and paste it into the extension sidebar.

Extension bridge (MCP)

For extension work, the slow path is edit → rebuild → manually click through Chrome → guess what broke. The extension bridge connects Cursor (or any MCP client) to your real unpacked Chrome profile over localhost so agents can inspect live DOM state, fill fields, advance apply steps, and run Auto Apply without you acting as the robot.

Typical wins:

See what the extension sees - get_field_inventory, read_field_values, get_page_html on the tab you already have open (cookies, SSO, Reed/Indeed session intact).
Step through apply flows - platform tools like reed_tab_message, indeed_tab_message, linkedin_tab_message call the same content-script handlers the orchestrator uses (REED_APPLY_STATE, REED_FILL_AND_ADVANCE, etc.).
Run Auto Apply from the agent - start_auto_apply + auto_apply_status / auto_apply_stop for LinkedIn, Indeed, Totaljobs, Glassdoor, or Reed.
Tight rebuild loop - npm run extension:build-reload rebuilds extension/dist/ and asks the connected extension to reload without a manual trip to chrome://extensions.

How it fits together

flowchart LR
    Cursor["Cursor agent<br/>MCP tools"]
    MCP["mcp-server.mjs<br/>stdio"]
    HTTP["Bridge HTTP<br/>127.0.0.1:7433"]
    WS["Bridge WebSocket<br/>127.0.0.1:7432"]
    BG["Extension<br/>service worker"]
    CS["Content scripts<br/>on job tabs"]

    Cursor --> MCP --> HTTP --> WS --> BG --> CS

Piece	Role
`extension/src/shared/bridge-client.js`	Outbound WebSocket from the MV3 background worker
`scripts/extension-bridge/server.mjs`	Local WebSocket + HTTP command API (`npm run extension-bridge`)
`scripts/extension-bridge/mcp-server.mjs`	MCP stdio wrapper over the HTTP API

The bridge is dev-only: unpacked builds from extension/dist/ auto-connect on localhost. It binds to 127.0.0.1 only.

Setup

Terminal 1 - bridge (leave running):

npm run extension-bridge

Terminal 2 - Laravel + Vite as usual (composer run dev).

Chrome - load unpacked from extension/dist/ (see Build the browser extension). Connect the extension to your dashboard token.

Cursor - register the MCP server in .cursor/mcp.json (use your repo path for cwd):

{
  "mcpServers": {
    "autocvapply-extension": {
      "command": "node",
      "args": ["scripts/extension-bridge/mcp-server.mjs"],
      "cwd": "/path/to/autocvapply"
    }
  }
}

Restart Cursor after editing MCP config. The bridge server must be running before MCP tools work.

Verify:

curl -s http://127.0.0.1:7433/status

Expect "extensionConnected": true when Chrome has the unpacked extension loaded.

Fast iteration loop

# After editing extension/src/** or form-heuristics:
npm run extension:build-reload

# Optional: targeted platform unit test (no browser)
node --test scripts/extension-test/reed-platform.test.mjs

# Optional: drive Auto Apply from Node (uses the same bridge HTTP API)
node scripts/extension-test/auto-apply-marathon.mjs --platform=reed --target=5 --role="software engineer" --location=London --fit=false

In Cursor, a productive apply-debug session usually looks like:

extension_status or list_tabs - find the apply tab.
set_active_tab - pin it for follow-up commands.
reed_tab_message / indeed_tab_message / … with REED_APPLY_STATE (or equivalent) - read step fingerprint, validation errors, review/submit flags.
get_field_inventory + apply_answer or start_draft_all - fill the current step.
Platform _FILL_AND_ADVANCE message - click Continue/Submit in the real page.
save_fixture - capture redacted HTML into the form corpus when you find a new edge case.

For multi-step Indeed-style flows: fill → click_control (Continue) → wait_for_tab → repeat.

MCP tools (summary)

Tool	Use when
`extension_status`	Check bridge + extension connection and active tab
`get_field_inventory` / `read_field_values`	See fillable refs and live DOM values
`apply_answer` / `start_draft_all`	Fill one field or run Draft All on the pinned tab
`click_control` / `click_ref`	Advance wizards (Continue, Next, etc.)
`navigate_tab` / `wait_for_tab`	Open jobs or wait for SPA navigation
`linkedin_tab_message`, `indeed_tab_message`, `totaljobs_tab_message`, `glassdoor_tab_message`, `reed_tab_message`	Call platform content-script handlers directly
`start_auto_apply`, `auto_apply_status`, `auto_apply_stop`	Run or monitor job-board Auto Apply
`save_fixture`	Add redacted HTML to `tests/fixtures/form-extraction/`

Full tool list, HTTP examples, env vars, and security notes: scripts/extension-bridge/README.md.

For the full job-board shipping workflow (MCP autofill → synthetic corpus → live marathon → commit), see docs/platform-automation-playbook.md.

Form corpus quality engineering

This is not a side-project extension with a handful of smoke tests. AutoCVApply ships with one of the most exhaustive form-autofill verification pipelines in the job-application tooling space - built because a single missed combobox on a Greenhouse form is a failed application.

The numbers

Metric	Count	Source
Form extraction scenarios	6,229	`tests/fixtures/form-extraction/manifest.json`
Vetted scenarios	3,942	same manifest (`status: vetted`; remainder pending review)
HTML fixtures	6,232	`tests/fixtures/form-extraction/html/`
Expected field-inventory snapshots	6,006	`tests/fixtures/form-extraction/expected/`
Scraped real apply pages (`web-*`)	1,761	same manifest
Job board synthetic (`syn-tj-500-`, `syn-gd-300-`, `syn-reed-300-`, `syn-sh-300-`, `syn-cvl-300-*`)	1,700	Totaljobs, Glassdoor, Reed, SimplyHired, CV-Library Auto Apply flows
Curated fill-verify scenarios	124 (70 JSDOM · 54 Playwright)	`tests/fixtures/form-extraction/fill-verify-curated.json`
Platform smoke scenarios	13 (+ 2 Ashby widget checks)	`fill-verify-smoke.json` · `run-ashby-*-playwright.mjs`
Extension E2E scenarios	103 (10 in CI)	`tests/fixtures/extension-e2e/e2e-scenarios.json`
Profile-mapping benchmark scenarios	190	`scripts/extension-benchmark/profile-mapping-corpus.json`
Answer-quality benchmark scenarios	124	`scripts/extension-benchmark/answer-quality-corpus.json`
PHPUnit test methods	437	`tests/*/Test.php`
Platform buckets in curated tier	20	`scripts/form-corpus/lib/curated-manifest.mjs`

The corpus blends 1,761 scraped real ATS and job-board pages (web-*) with 4,156 synthetic scenarios - including 500 syn-complex-500 ATS-style fixtures, 500 syn-corpus2 bulk forms, 500 syn-tj-500 Totaljobs flows, 300 syn-gd-300 Glassdoor flows, 300 syn-reed-300 Reed flows, 300 syn-sh-300 SimplyHired flows, 300 syn-cvl-300 CV-Library flows, 60 syn-weird edge cases, plus framework mega-forms for React, Vue, Angular, Svelte, Shadow DOM, Workday wizards, conditional fields, and combobox edge cases.

The test pyramid

Every change to form-heuristics.js or field-inventory.js must survive the full pyramid before merge:

flowchart TB
    subgraph L1["Layer 1 - Fast feedback"]
        U["PHPUnit unit tests<br/>437 methods"]
        P["Propagation + mock answer tests"]
        D["Debug log golden replay"]
    end

    subgraph L2["Layer 2 - Curated JSDOM"]
        J["70 synthetic scenarios<br/>4-layer verification · 100% pass"]
    end

    subgraph L3["Layer 3 - Real browser"]
        S["Platform smoke<br/>13 ATS fixtures + Ashby widgets"]
        PW["Curated Playwright<br/>54 scraped ATS pages"]
        VR["Visual regression<br/>screenshot baselines"]
    end

    subgraph L4["Layer 4 - Full extension"]
        E2E["Extension E2E<br/>103 scenarios · real MV3 + mocked API"]
    end

    L1 --> L2 --> L3 --> L4

Tier	Engine	Scope	CI job
Unit	JSDOM / Node	Propagation, mock answers, debug-log replay	`tests.yml` → `php-tests`
Corpus sanity	PHPUnit	Corpus size checks (6,229 scenarios, 3,942 vetted)	`tests.yml` → `php-tests`
Form extraction eval	JSDOM	All 3,942 vetted scenarios vs expected field inventory	`tests-heavy.yml` (manual)
Curated JSDOM	JSDOM	70 synthetic scenarios, 4-layer checks at 100%	`tests.yml` → `extension-fill`
Platform smoke	Playwright	1 scenario per ATS/platform + Ashby yes/no + checkbox	`tests.yml` → `extension-fill`
Curated Playwright	Playwright	54 priority scraped ATS fixtures	`tests-heavy.yml` (manual)
Visual regression	Playwright + pixelmatch	Baseline compare on smoke subset	`tests-heavy.yml` (manual)
Extension E2E	Playwright + unpacked MV3	Full Draft All with mocked assist API (103 total · 10 in CI)	`extension-fill` (CI subset) · `tests-heavy.yml` (full batch)

Four layers of fill verification

Each curated JSDOM scenario passes four independent checks - not just "did we set a value?":

Layer	What it proves
DOM readback	Re-reads filled values from the DOM after `applyAnswerByRefAllFrames`
HTML5 validity	`element.checkValidity()` / `form.checkValidity()` on native controls
Accessibility state	`aria-checked`, `aria-selected`, `aria-pressed`, combobox collapsed state
Error banners	Ashby/Greenhouse-style validation messages, `[role="alert"]`, `[aria-invalid="true"]`

Additional tiers add OCR readback (Playwright + Tesseract on Ashby fixtures), pixel diff (before/after screenshot % change), and debug log golden replay (extension phase summaries).

Pass-rate thresholds (enforced in CI)

Tier	Critical	Overall
JSDOM curated	100%	100%
Playwright priority	100%	100%
Platform smoke	100%	100%
Extension E2E	100%	100%

CI pipeline

Fast feedback runs on every push to main and develop; the heavy corpus eval tier is manual-only:

Workflow	Trigger	What runs
`tests.yml` → `php-tests`	Push / PR	Laravel suite on PostgreSQL 17 - excludes `@group playwright` and `@group extension-e2e`; corpus sanity checks only (not full extraction eval)
`tests.yml` → `extension-fill`	Push / PR	`npm run build:extension` → curated JSDOM verify (70 scenarios) → Playwright platform smoke
`tests-heavy.yml`	Manual dispatch	Full 3,942-scenario extraction eval · comprehensive fill verify · curated Playwright · visual regression · extension E2E batch (~103 scenarios)
`lint.yml`	Push / PR	Laravel Pint, ESLint, Prettier
`prod_deploy.yml`	Push to `main`	Docker build → GHCR push → deploy to production

After changing form heuristics locally, run the smoke tier before opening a PR:

npm run form-corpus:fill-verify:smoke
# or
FORM_CORPUS_PLAYWRIGHT=1 php artisan test --compact --filter=test_platform_smoke_playwright_passes

See scripts/form-corpus/README.md for the full maintenance workflow, report paths, and the manual heavy tier.

Job board Auto Apply testing

LinkedIn, Indeed, Totaljobs, Glassdoor, and Reed Auto Apply have their own verification pyramid, separate from the 6,229-scenario ATS form corpus. It runs on every push via PHPUnit wrappers and can be exercised locally without a live session for most tiers.

flowchart TB
    subgraph L1["Layer 1 - Unit (JSDOM / Node)"]
        U1["linkedin-auto-apply.mjs<br/>search URLs, job cards, modal state"]
        U2["auto-apply-pause-resume.mjs<br/>blockers + session pause/resume"]
        U3["auto-apply-activity-visibility.mjs<br/>sidebar activity UI rules"]
        U4["linkedin-full-flow-report.mjs<br/>E2E report parsing"]
    end

    subgraph L2["Layer 2 - Offline corpus (JSDOM)"]
        C["linkedin-easy-apply-corpus.mjs<br/>212 live captures + 52 synthetic edge cases"]
        I["indeed-auto-apply-offline.mjs<br/>44 apply-step fixtures"]
        TJ["syn-tj-500-* · syn-gd-300-* · syn-reed-300-* · syn-sh-300-* · syn-cvl-300-*<br/>1,700 job board synthetic scenarios"]
    end

    subgraph L3["Layer 3 - Browser offline"]
        O["linkedin-auto-apply-offline-step.mjs<br/>Playwright + fixture HTML"]
    end

    subgraph L4["Layer 4 - Live E2E (manual)"]
        E["linkedin-auto-apply-full-flow.mjs<br/>real LinkedIn session + MV3 extension"]
    end

    L1 --> L2 --> L3 --> L4

Tier	Script / test	What it validates	CI
Unit	`scripts/extension-test/linkedin-auto-apply.mjs`	Search URL building, job card parsing, Easy Apply button state, modal detection, step fingerprints, cookie/save-dialog dismiss	`LinkedInAutoApplyTest`
Unit	`auto-apply-pause-resume.mjs`, `auto-apply-activity-visibility.mjs`	Validation blockers, pause/resume session state, sidebar activity panel visibility	Default PHPUnit
Offline corpus	`linkedin-easy-apply-corpus.mjs`	Each fixture: modal open/closed, primary actions (Next/Review/Submit), validation errors, submitted confirmation, multi-step flow progression	`LinkedInEasyApplyCorpusTest`
Offline corpus	`indeed-auto-apply-offline.mjs`, `indeed-apply-*.mjs`	Indeed Apply search URLs, step detection, contact/resume/questions/review fixtures	`npm run test:indeed-auto-apply-offline`
Offline corpus	`syn-tj-500-`, `syn-gd-300-`, `syn-reed-300-`, `syn-sh-300-`, `syn-cvl-300-*`	1,700 synthetic Totaljobs, Glassdoor, Reed, SimplyHired, and CV-Library search/apply scenarios in the form corpus	`npm run form-corpus:validate-*-corpus`
Unit	`reed-platform.test.mjs`	Reed search URLs, job ID parsing, apply URL builders	`node --test scripts/extension-test/reed-platform.test.mjs`
Browser offline	`linkedin-auto-apply-offline-step.mjs`	Real Chrome + unpacked extension clicking through a fixture modal step	`LinkedInFullFlowReportTest` (requires `EXTENSION_E2E=1`)
Live E2E	`linkedin-auto-apply-full-flow.mjs`	Full orchestrator: search, open jobs, Draft All per step, advance, submit; writes `tests/output/linkedin-auto-apply-full-flow/report.json`	Manual (`LINKEDIN_LIVE_E2E=1`)

Fixtures

Location	Contents
`tests/fixtures/auto-apply/linkedin/captured/`	212 live-captured LinkedIn HTML files (authoritative regression source)
`tests/fixtures/form-extraction/html/web-indeed-*`	44 Indeed Apply step fixtures (contact, resume, questions, review)
`tests/fixtures/form-extraction/html/syn-tj-500-*`	500 Totaljobs Quick Apply synthetic scenarios
`tests/fixtures/form-extraction/html/syn-gd-300-*`	300 Glassdoor Easy Apply host + iframe scenarios
`tests/fixtures/form-extraction/html/syn-reed-300-*`	300 Reed Easy Apply search, job detail, and modal review scenarios
`tests/fixtures/auto-apply/linkedin/captured-manifest.json`	Metadata per capture: step, `capture_reason`, validation errors, stuck diagnostics
`tests/fixtures/auto-apply/linkedin/`	50 synthetic progression flows plus `error-` and `edge-` fixtures (fallback when captures are unavailable)
`tests/fixtures/auto-apply/linkedin-search-results*.html`	Search results and job detail page fixtures for card parsing

See tests/fixtures/auto-apply/linkedin/README.md for capture flags, stuck-step suffixes, and fixture layout.

Commands

Run the fast tiers before changing linkedin-auto-apply.js, linkedin-easy-apply-fields.js, or the auto-apply orchestrator:

# Unit tests (search, cards, modal state)
node scripts/extension-test/linkedin-auto-apply.mjs

# Pause/resume blockers and sidebar activity UI
npm run test:auto-apply-pause-resume
npm run test:auto-apply-activity-visibility

# Offline corpus (CI default - prefers live captures)
npm run test:linkedin-easy-apply-corpus:run -- --captured-only

# Full corpus including synthetic fallback
npm run test:linkedin-easy-apply-corpus

# PHPUnit wrappers (runs Node scripts above)
php artisan test --compact tests/Unit/Extension/LinkedInAutoApplyTest.php
php artisan test --compact tests/Unit/Extension/LinkedInEasyApplyCorpusTest.php

Browser offline step (build extension first):

npm run build:extension
npm run test:linkedin-full-flow:offline-step
# or via PHPUnit:
EXTENSION_E2E=1 php artisan test --compact --filter=test_linkedin_auto_apply_offline_step

Live full-flow E2E (requires LinkedIn test account in .env):

npm run build:extension
LINKEDIN_LIVE_E2E=1 npm run test:linkedin-full-flow -- --max-jobs=3 --roles="software engineer"
# or via PHPUnit:
LINKEDIN_LIVE_E2E=1 php artisan test --compact tests/Feature/Extension/LinkedInAutoApplyFullFlowTest.php

Indeed Apply fast tiers

Run before changing indeed-auto-apply.js, indeed-platform.js, or the auto-apply orchestrator:

# Offline orchestrator + step fixtures (contact, resume, questions, review)
npm run test:indeed-auto-apply-offline
npm run test:indeed-apply-contact
npm run test:indeed-apply-questions
npm run test:indeed-apply-job-006
npm run test:indeed-apply-job-007
npm run test:indeed-apply-job-008

Indeed step fixtures live under tests/fixtures/form-extraction/html/web-indeed-* and in the form-extraction manifest.

Adding live captures

Use a headed Playwright session to grow the offline corpus from real LinkedIn Easy Apply modals:

# Credentials in .env only (never commit):
# LINKEDIN_TEST_EMAIL=...
# LINKEDIN_TEST_PASSWORD=...

npm run extension-e2e:capture-linkedin-corpus -- --target-fixtures=50 --max-jobs=20 --roles="software engineer,frontend developer"

The capture script logs in, rotates search roles, opens Easy Apply jobs, saves step states (open, filled, validation errors, multi-step review, stuck flows), redacts PII before write, and updates captured-manifest.json.

After capture or sanitizer rule changes, re-process fixtures before committing:

node scripts/extension-e2e/resanitize-linkedin-captured.mjs
node scripts/extension-e2e/rebuild-captured-manifest.mjs
npm run test:linkedin-easy-apply-corpus:run -- --captured-only

PII redaction and secret checks

Captured HTML is sanitized on save and can be re-sanitized in bulk:

Strips <script> tags
Replaces emails with candidate@example.com, phone with +44 7700 900123, names with Alex Candidate
Scrubs .env credentials (LINKEDIN_TEST_EMAIL, LINKEDIN_TEST_PASSWORD) and known API key patterns via redactSecrets

Before committing new fixtures:

npm run secrets:check-fixtures

Do not paste unredacted live LinkedIn HTML into tests/fixtures/auto-apply/linkedin/captured/.

Key commands

Command	Purpose
`composer run dev`	Laravel + queue + Pail + Vite
`composer test`	Pint check + full PHPUnit suite
`npm run build:extension`	Build MV3 extension to `extension/dist/`
`npm run extension:build-reload`	Build + hot-reload unpacked extension via bridge
`npm run extension-bridge`	Start localhost bridge (WebSocket + HTTP for MCP)
`npm run form-corpus:fill-verify:curated`	Curated JSDOM tier (CI default)
`npm run form-corpus:fill-verify:smoke`	Per-platform Playwright smoke
`npm run platform-logos:download`	Refresh platform logos in `public/images/platforms/logos/`
`npm run readme:platforms`	Regenerate Supported platforms block in README from `site.ts`
`npm run form-corpus:visual-regression`	Screenshot baseline compare
`npm run form-corpus:build-curated`	Regenerate curated + smoke manifests
`npm run test:linkedin-easy-apply-corpus:run -- --captured-only`	LinkedIn offline corpus (212 live captures)
`npm run test:linkedin-full-flow`	LinkedIn live E2E (requires `LINKEDIN_LIVE_E2E=1` + credentials)
`npm run extension-e2e:capture-linkedin-corpus`	Capture live LinkedIn Easy Apply HTML for offline tests
`npm run lint:check`	ESLint
`composer lint:check`	Laravel Pint

PHPUnit tiers

# Default CI (excludes playwright + extension-e2e groups)
php artisan test --compact --exclude-group=extension-e2e,playwright

# Playwright smoke (fast CI tier)
FORM_CORPUS_PLAYWRIGHT=1 php artisan test --compact --group=playwright --exclude-group=extension-e2e

# Extension E2E CI subset
EXTENSION_E2E=1 php artisan test --compact --group=extension-e2e

# Full ~103 scenario extension E2E (manual via tests-heavy.yml, 30-60+ min)
EXTENSION_E2E=1 EXTENSION_E2E_FULL=1 php artisan test --compact --group=extension-e2e

API

The extension authenticates with Laravel Sanctum bearer tokens.

Method	Endpoint	Description
`GET`	`/analytics.json`	Public aggregate usage totals (no auth; for README badges)
`GET`	`/api/profile`	Fetch user profile + subscription usage
`POST`	`/extension/connection`	Generate extension connection JSON (dashboard session)
`POST`	`/api/applications/assist/inventory`	Field inventory for current page
`POST`	`/api/applications/assist/job-context`	Extract job title, company, description
`POST`	`/api/applications/assist/draft-all`	Stream batch field answers (NDJSON)
`POST`	`/api/applications/assist/draft-field`	Single field draft
`POST`	`/api/applications/assist/cover-letter`	Generate cover letter
`POST`	`/api/applications/assist/tailored-resume`	Tailored resume draft
`DELETE`	`/api/tokens/{token}`	Revoke a token

Deployment

Production runs in Docker (DockerfileProd) with Nginx, PHP-FPM, a queue worker, and schedule:work (via supervisord). Pushes to main build a GHCR image and deploy via GitHub Actions. The Laravel scheduler must keep running for weekly blog:generate (Mondays 09:00 app timezone / UTC) and the worker heartbeat job.

CV extraction and Draft All use deepseek/deepseek-v4-flash:throughput by default (config/cv.php). To override the model in production, set NANOGPT_CV_MODEL in the server .env at /opt/autocvapply/.env; remove that key to use the config default after deploy.

Live site: autocvapply.com

Contributing

Issues and pull requests welcome on GitHub.

Fork the repo
Create a feature branch
Write tests for your changes
If you touched form-heuristics.js or field-inventory.js, run npm run form-corpus:fill-verify:smoke
Run composer test and npm run lint:check
Open a PR

License

PolyForm Noncommercial License 1.0.0. Free for personal and non-commercial use - fork it, study it, run it locally. Commercial use (selling, SaaS, paid services built on this code, etc.) requires permission; contact us via autocvapply.com/contact.

Get started free at autocvapply.com
_{Built for people who'd rather apply to jobs than retype their CV.

Verified against 6,229 form scenarios. Battle-tested on real ATS platforms and UK job boards.}

Name		Name	Last commit message	Last commit date
Latest commit History 438 Commits
.agents/skills		.agents/skills
.cursor		.cursor
.gemini		.gemini
.githooks		.githooks
.github/workflows		.github/workflows
app		app
bootstrap		bootstrap
config		config
database		database
docker		docker
docs		docs
extension		extension
public		public
resources		resources
routes		routes
scripts		scripts
storage		storage
tests		tests
.editorconfig		.editorconfig
.env.example		.env.example
.gitattributes		.gitattributes
.gitignore		.gitignore
.node-version		.node-version
.npmrc		.npmrc
.prettierignore		.prettierignore
.prettierrc		.prettierrc
AGENTS.md		AGENTS.md
Dockerfile		Dockerfile
DockerfileProd		DockerfileProd
GEMINI.md		GEMINI.md
LICENSE		LICENSE
README.md		README.md
artisan		artisan
boost.json		boost.json
components.json		components.json
compose.prod.yaml		compose.prod.yaml
compose.yaml		compose.yaml
composer.json		composer.json
composer.lock		composer.lock
eslint.config.js		eslint.config.js
package-lock.json		package-lock.json
package.json		package.json
phpunit.xml		phpunit.xml
pint.json		pint.json
pnpm-workspace.yaml		pnpm-workspace.yaml
tsconfig.json		tsconfig.json
vite.config.ts		vite.config.ts

Folders and files

Latest commit

History

Repository files navigation

AutoCVApply

Table of contents

See it in action

What is AutoCVApply?

Without vs with AutoCVApply

Who it's for

Quick install

Production (autocvapply.com)

From source (developers)

How it works

Job board Auto Apply

Features

Cover letters

Human-sounding AI drafts

Anti-bot detection

Supported platforms

Autofill on ATS and career sites

Auto Apply - supported today

Auto Apply - coming soon

Fill verification depth (ATS)

Postbox design

Pricing

Security & privacy

Testing overview

Job search tips

Links

Architecture

Tech stack

Project structure

Getting started (local dev)

Prerequisites

Install

Environment

Run locally

Build the browser extension

Extension bridge (MCP)

How it fits together

Setup

Fast iteration loop

MCP tools (summary)

Form corpus quality engineering

The numbers

The test pyramid

Four layers of fill verification

Pass-rate thresholds (enforced in CI)

CI pipeline

Job board Auto Apply testing

Fixtures

Commands

Indeed Apply fast tiers

Adding live captures

PII redaction and secret checks

Key commands

PHPUnit tiers

API

Deployment

Contributing

License

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages