khetansarvesh/job-apply

Job Application Filler

Automate job application filling using Chrome DevTools MCP tools. Uses the Simplify browser extension for initial auto-fill, then intelligently fills remaining fields from the user's resume and project descriptions.

When to Activate

• User asks to fill/apply to job applications
• User says "apply to all evaluated jobs" or "apply to everything"

Prerequisites

• Chrome browser open with Chrome DevTools MCP connected
• Simplify browser extension installed in Chrome
• Profile data filled in:
  • Notion (fetch via python3 scripts/notion/page_reader.py resume) — personal info, education, experience, skills (shared, at repo root)
  • Notion (fetch via python3 scripts/notion/page_reader.py projects) — project descriptions and reusable answer snippets (shared, at repo root)
• (Optional) Prior evaluation from job-eval — if the role was evaluated, the skill uses the evaluation report for smarter, archetype-aware answer generation
• NOTION_TOKEN environment variable set — for querying evaluated jobs and updating status

Non-Negotiables

1. NEVER click Submit/Apply buttons — always stop before submission and ask the user to review
2. NEVER auto-fill sensitive fields (SSN, bank details, salary if user hasn't specified) — flag these for the user
3. Always take a final screenshot for user review before declaring done
4. Read profile data fresh every time — never rely on cached/remembered values
5. ALWAYS trigger Simplify FIRST — finding and clicking the Simplify autofill button is the #1 priority before any manual filling. Try multiple methods (screenshot, snapshot, evaluate_script) to locate it. Only skip if Simplify is genuinely not installed.
6. Cover letter handling:
   • If the form marks cover letter as optional → leave empty (don't upload unless user asks)
   • If the form marks cover letter as required → generate one:
     • 1 page max, 3-4 paragraphs
     • Paragraph 1: Why this company specifically (concrete fact, not flattery)
     • Paragraph 2: Your strongest match point with a quantified result
     • Paragraph 3: Bridge — how your background connects to what they need
     • Paragraph 4: Forward-looking close
     • Use JD keywords naturally throughout
     • Save as text in the textarea, or generate PDF if file upload required

Workflow

Step 1: Get Evaluated Jobs from Notion

Run the query script to get all "Evaluated" jobs from Notion:

python3 scripts/notion/db_applications.py query --status "Evaluated"

Parse the JSON output — get list of {page_id, company, role, url, score}.

Filter: only process jobs that have a URL (skip rows with url: null).

Show the user the list and ask for confirmation:

Found N evaluated jobs to apply to:
1. Cohere — Applied AI Engineer (4.4) — https://...
2. Together AI — Research Engineer (3.8) — https://...
...
Proceed with all N?

Step 2: Load Profile Data & Evaluation Context

Read profile files to have all user data available:

• Read Notion (fetch via python3 scripts/notion/page_reader.py resume) for structured data (name, email, education, experience, skills)
• Read Notion (fetch via python3 scripts/notion/page_reader.py projects) for project summaries and reusable answer snippets
• For open-ended questions requiring detailed answers, fetch project context from Notion:

  python3 scripts/notion/page_reader.py roma          # ROMA recursive deep search agent
  python3 scripts/notion/page_reader.py sera          # SERA semantic embedding agent
  python3 scripts/notion/page_reader.py mroma         # m-ROMA multimodal extension
  python3 scripts/notion/page_reader.py deep-research # Multi-agent deep research at Strategy
  python3 scripts/notion/page_reader.py txt2sql       # Text2SQL agent at Piramal
  

  Requires NOTION_TOKEN env var set.

If profile files are empty/incomplete, warn the user and ask them to fill in the data first.

Load evaluation context (if available):

Each row in the Notion database is a page. The evaluation report is stored as the page body content. Fetch it using the existing script with the page_id directly:

# The page_id comes from db_applications.py query output
python3 scripts/notion/page_reader.py {page_id}

If the page body has content, extract:

• Archetype — determines answer framing (Research Engineer answers differ from Applied AI answers)
• Block B gaps — know what gaps to proactively address in open-ended answers
• Block F STAR stories — pre-mapped stories ready to use for behavioral questions
• Keywords — extracted JD keywords to weave into answers

When evaluation context is available, use these enhanced framing rules for open-ended answers:

• Tone: "I'm choosing you" — frame answers as selective ("What drew me to [Company] is..."), never desperate
• Archetype-aware: Research/ML roles → lead with SOTA results, publications. Applied/Platform roles → lead with production systems, scale metrics
• Use evaluation proof points: For "Why this role?" → reference matching proof point. For "Relevant achievement?" → use a pre-mapped STAR story

If no evaluation exists, proceed with profile data only — the skill works fine without it.

Step 3: Process Each Job

For each job, sequentially:

3a. Open in New Tab

ALWAYS open each job in a NEW TAB — use mcp__chrome-devtools__new_page (NEVER navigate in existing tab).

3b. Trigger Simplify Extension

Simplify adds a floating button overlay to job application pages. Find and click it:

1. Take a screenshot to locate the Simplify button

   • Look for a blue/green floating button, often in the bottom-right corner
   • It may say "Autofill" or show the Simplify logo
   • Common selectors: [class*="simplify"], [id*="simplify"], button with Simplify text

2. Click the Simplify button:

   Use mcp__chrome-devtools__click with the Simplify button's location
   

3. Wait for Simplify to populate fields:

   Use mcp__chrome-devtools__wait_for with timeout of 5000ms for form fields to be filled
   

4. If Simplify button is not found, skip to Step 3c (manual fill all fields)

5. If Simplify opens a modal/popup, interact with it to confirm auto-fill

3c. Screenshot & Analyze Gaps

After Simplify runs (or if skipped):

1. Take a screenshot of the full form

2. Scroll down and take additional screenshots if the form is long

   Use mcp__chrome-devtools__press_key with "PageDown" then screenshot again
   Repeat until you've seen the entire form
   

3. Identify unfilled fields by analyzing the screenshots:

   • Empty text inputs (no value visible)
   • Unselected dropdowns (showing "Select..." or placeholder)
   • Unchecked required checkboxes
   • Empty textareas (especially for open-ended questions)
   • File upload fields without files attached

3d. Fill Remaining Fields

For each unfilled field, match it to profile data and fill:

Text Inputs (name, email, phone, LinkedIn, etc.)

Use mcp__chrome-devtools__click on the field first
Use mcp__chrome-devtools__fill with the value from resume.md

Dropdowns (work authorization, degree type, etc.)

Use mcp__chrome-devtools__click to open the dropdown
Use mcp__chrome-devtools__take_screenshot to see options
Use mcp__chrome-devtools__click on the correct option

Textareas (cover letter, "why this role", project descriptions)

Use mcp__chrome-devtools__click on the textarea
Use mcp__chrome-devtools__fill with content adapted from projects.md

For open-ended questions, follow these strict rules:

• Plain simple English — no jargon-heavy sentences, no filler words, no fluff
• Under 150 words — be short, crisp, and impactful. Every sentence must earn its place
• Lead with the result — start with what you achieved, then briefly explain how
• One project per answer — pick the single most relevant project from projects.md, don't cram multiple
• Use real numbers — percentages, dollar amounts, team sizes, time saved
• No generic statements — every answer must reference a specific project, tool, or metric from the profile
• Read the question carefully — answer exactly what's asked, nothing more
• Fetch detailed project context from Notion (python3 scripts/notion/page_reader.py {project}) when composing answers

File Uploads (resume PDF, cover letter)

Use mcp__chrome-devtools__upload_file with the path from resume.md "Resume PDF Path" field

Checkboxes / Radio Buttons

Use mcp__chrome-devtools__click on the appropriate option

Field Matching Guide

| Form Label Pattern | Profile Source | | ---------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------- | | First Name, Last Name, Full Name | resume.md > Personal Information > Full Name | | Email, E-mail | resume.md > Personal Information > Email | | Phone, Mobile, Telephone | resume.md > Personal Information > Phone | | LinkedIn, LinkedIn URL | resume.md > Personal Information > LinkedIn | | GitHub, Portfolio, Website | resume.md > Personal Information > GitHub/Portfolio | | City, Location, Address | resume.md > Personal Information > Location | | University, School, College | resume.md > Education > University | | Degree, Education Level | resume.md > Education > Degree | | Major, Field of Study | resume.md > Education > Major | | GPA | resume.md > Education > GPA | | Graduation Date, Grad Year | resume.md > Education > Graduation Date | | Current Company, Employer | resume.md > Work Experience > Experience 1 > Company | | Current Title, Job Title | resume.md > Work Experience > Experience 1 > Title | | Years of Experience | resume.md > Additional Information > Years of Experience | | Work Authorization, Legally authorized | resume.md > Work Authorization | | Visa Sponsorship, Require sponsorship | resume.md > Work Authorization > Visa sponsorship required | | Salary, Compensation, Pay | Flag for user -- do not auto-fill unless specified in resume.md | | Start Date, Available from | resume.md > Additional Information > Earliest Start Date | | How did you hear about us, Referral | resume.md > Additional Information > How did you hear about us | | Gender, Race, Ethnicity, Veteran, Disability | resume.md > Additional Information (EEO fields -- use values if provided, otherwise select "Decline to self-identify") | | Resume, CV (file upload) | resume.md > Resume PDF Path | | Cover Letter (file upload) | resume.md > Cover Letter PDF Path | | Tell us about yourself, Why interested, Describe a project | projects.md > select most relevant snippet and adapt |

3e. Handle Platform-Specific Quirks

Greenhouse

• Multi-page forms: after filling visible fields, look for "Next" or "Continue" button
• Click it and repeat Steps 3c-3d for the next page
• Custom questions often appear on page 2+

Lever

• Single-page forms with sections
• Resume upload is usually at the top
• Custom questions at the bottom

Workday

• Complex multi-step wizard
• Requires account creation sometimes -- flag for user
• Many dropdown-heavy fields

Ashby

• Clean single-page layout
• File upload then form fields

Generic ATS / Custom Forms

• Use screenshot analysis to identify form structure
• Match labels to profile data using the field matching guide above

3f. Take Final Screenshot

Take a screenshot of the completed form for reference.

3g. Update Notion Status

Update the row to "Almost Applied":

python3 scripts/notion/db_applications.py update-status --page-id "{page_id}" --status "Almost Applied"

3h. Move to Next Job

Do NOT wait for user to submit — move on immediately to the next job.

Step 4: Summary

After all jobs processed, summarize:

Applied to N jobs. All forms filled and open in separate tabs.
Review each tab and click Submit when ready.

Tabs open:
1. Cohere — Applied AI Engineer → Status: Almost Applied
2. Together AI — Research Engineer → Status: Almost Applied
...

Critical Rules

• ALWAYS open each job in a NEW TAB — never navigate away from a tab with a filled form
• NEVER wait for user to submit before moving to the next job
• NEVER click Submit — user reviews and submits each tab manually
• Update Notion status after filling each form — not at the end

Handling Edge Cases

CAPTCHA

• Flag for user: "There's a CAPTCHA on this form. Please solve it manually, then tell me to continue."

Login Required

• Flag for user: "This application requires login. Please log in first, then tell me to continue."

Multi-page Forms

• Fill each page completely before advancing
• Take screenshots of each page for the review summary

Dynamic Fields (fields that appear based on previous answers)

• After filling a field that might trigger new fields, take a screenshot
• If new fields appeared, fill those too

Pre-filled but Incorrect Fields

• If Simplify filled a field incorrectly, clear it first:

  Use mcp__chrome-devtools__click on the field (triple-click to select all)
  Use mcp__chrome-devtools__fill with the correct value
  

Troubleshooting

| Issue | Solution | | ----------------------------- | --------------------------------------------------------------------------- | | Simplify button not found | Skip Simplify, fill all fields manually from profile data | | Chrome DevTools not connected | Ask user to ensure Chrome is open and MCP server is running | | Field won't accept input | Try clicking the field first, then use type_text instead of fill | | Dropdown options not visible | Take screenshot after clicking, use press_key with arrow keys to navigate | | Form has iframes | Use evaluate_script to check for iframes, switch context if needed | | File upload fails | Verify the PDF path in resume.md is correct and file exists |

Related Skills

• job-eval — Evaluate a job offer before applying (A-G scoring framework)
• job-interview-prep — Prepare for interviews at a specific company
• job-cv-tailor — Generate an ATS-optimized CV tailored to the JD
• job-ops — Command center that routes to all career skills
• linkedin-outreach — Draft personalized LinkedIn outreach messages