arrow_backBack to Projects
robot_2

pick me.

Agent Plugin
Oct 16, 2026visibility15

A personal job application agent for Google Antigravity that discovers verified job openings, tailors single-column resumes, drafts authentic cover letters, and tracks applications.

Google AntigravityTypstPowerShellPythonPlaywrightYAML

Installation via Antigravity CLI:

agy plugin install https://github.com/auxghlann/pick-me

A straightforward breakdown of what this project is, the problem it solves, and how it works.

What is Pick Me?
image.pngzoom_in Pick Me is a career assistance suite built as an agent plugin (right now, it only available for antigravity). It is structured around two distinct entities: a set of inline document builders that tailor single-column Typst resumes, write authentic cover letters, and manage your application pipeline in a local CSV, and a dedicated crawler subagent that independently discovers, scores, and live-verifies tech job openings matching your real years of experience,.


1. Why I Built This

As a fresh Computer Science graduate looking for jobs (for months now), the job hunt is kinda feeling repetitive and exhausting process; soul-sucking even.

A typical day usually involves repeating the same manual loop dozens of times:

  • Searching job boards for entry-level roles, only to find that most "junior" listings actually demand 4 to 6 years of experience and don't even get me started on their recommendation system. (just why??? these big job posting companies can't even write a good recommendation system).
  • Tweaking resume bullet points to match the exact wording of each job description.
  • Writing a personalized cover letter from scratch for each company.
  • Keeping track of application in a spreadsheet.

Other than that, turning to general AI tools like Gemini or ChatGPT often created new problems rather than saving time. When asked to draft a resume or cover letter, LLMs tend to hallucinate, inventing tools I never used, exaggerating responsibilities that isn't even there, or making up metrics that I would never be able to defend honestly in a technical interview.

I wanted a tool that would respect my actual background, write in a genuine human voice, filter out jobs I am not qualified for yet, and produce clean documents without extra overhead.

And it got me thinking, what if I create one? I can imagine how to do it, and I know the exact tools to be used because I've been messing around on creating my very own Subagent, SKILLs or Plugins for my agent.

And then, after a day of thinking and arguing with my AI on what to do and how to do it, pick-me was born!


2. What It Does: Core Features

Pick Me is organized into two distinct parts: an independent job discovery crawler and a set of inline document builders, all anchored to a persistent candidate profile.

Persistent Profile & STAR Stories (The Source of Truth)

The foundation of the entire suite rests on two persistent local files: data/candidate-profile.yaml and data/star-stories.md.

  • Why This Matters: Standard AI models easily hallucinate when asked to tailor a resume or write a cover letter. If a job posting asks for specific frameworks or tools, an ungrounded model will often invent past experience or make up fake metrics to appear like a perfect match. That is dangerous in a technical interview where you have to explain and defend every bullet point on your resume.
  • User-Authored Ground Truth: By creating and maintaining your profile facts and STAR stories yourself, you give the AI an unyielding boundary. The AI is strictly restricted to your real projects, verified skills, and genuine metrics. If a skill or metric does not exist in your profile, the AI cannot invent it.
  • Cross-Session Memory: Instead of pasting your resume into a chat prompt over and over every time you tailor an application, your background is stored locally once and persists cleanly across sessions.

Independent Market Scout (job-posting-crawler)

The crawler functions as an isolated scout that searches top ATS boards (e.g, Workable, Greenhouse, Lever, Ashby) and regional portals (e.g., JobStreet, Indeed, Linkedin, Kalibrr):

  • Years of Experience Gate: Evaluates roles using a 100-point rubric that dedicates 40 points to experience level. If a posting requires 3 or more years beyond my actual experience, it is automatically discarded so I do not waste time applying to senior positions.
  • Freshness Filter: Filters out any job posting published more than 3 days (72 hours) ago to maximize early application response rates.
  • Live Status Verification: Checks the listing URL directly via HTTP requests (or Playwright browser automation for JavaScript-rendered portals) to confirm the job is actively accepting applications before recommending it.

Grounded Resume Builder (resume-builder)

  • Single-Column ATS Layout: Uses a clean, single-column Typst template that ATS text parsers can read from top to bottom without column-jumping errors.
  • Profile-Grounded Content: Reorders and prioritizes relevant coursework, certifications, and technical projects from data/candidate-profile.yaml to match the target role, while strictly refusing to invent missing skills or fake achievements.

Authentic Cover Letter Builder (cover-letter-builder)

  • Direct and Brief: Drafts brief 3 paragraph cover letters with full Typst text justification.
  • Anti-Buzzword Standards: Avoids generic openers and artificial hype, focusing directly on the specific problem the company is solving and connecting it to real projects from my portfolio.

Application Pipeline Tracker (application-tracker)

  • Centralized CSV Ledger: Maintains application records in data/applications.csv following RFC 4180 standards.
  • Automatic Target Dates: Assigns sequential application IDs (APP-001, APP-002) and schedules follow-up check-ins between 7 and 14 days after submission.

Human-in-the-Loop Review Gate

Before any .typ source file is written or compiled, Pick Me pauses and presents the complete 1:1 drafted text for review. Nothing is generated or modified without explicit user approval.


3. Optional Add-On Skills: Interview Prep & Well-Being

Beyond the core pipeline of finding jobs, tailoring documents, and logging applications, I built three optional modular skills to support the more human sides of the job search:

Interactive Interview Coach (interview-helper)

Preparing for technical and behavioral interviews is often stressful when practicing in isolation. The interview-helper skill serves as a private practice partner:

  • STAR Answer Structuring: Uses the candidate's real anecdotes from data/star-stories.md to help structure authentic answers across the Situation, Task, Action, and Result dimensions.
  • One-at-a-Time Mock Interviews: Simulates realistic interview pacing by asking one question at a time and evaluating responses against clear evaluation criteria instead of overwhelming the applicant with a giant wall of questions.
  • Reverse-Interview Questions: Provides tailored, high-signal questions for the candidate to ask hiring managers at the end of an interview, focusing on team deployment cadence, code review culture, and engineering debt.

Applicant Well-Being Coach (well-being-coach)

Job searching involves high levels of silence and automated rejection emails that can quickly lead to burnout. The well-being-coach provides grounded support:

  • Reframing Automated Rejections: Helps depersonalize automated ATS rejections without resorting to toxic positivity, treating applications as statistical iterations.
  • Sustainable Daily Pacing: Actively discourages demoralizing "spray-and-pray" blasts (submitting 50 generic resumes a day), encouraging a steady, sustainable pace of 2 to 3 well-tailored applications per day.
  • Pre-Interview Grounding: Provides simple perspective resets and breathing prompts before interview calls to help reduce anxiety.
  • Tracking Controllable Wins: Pulls stats from data/applications.csv to highlight positive momentum (such as total applications tailored or follow-ups completed) rather than focusing solely on callbacks.

Candidate Profile Onboarding (get-to-know-me)

  • An optional onboarding helper that extracts facts from an existing PDF or Markdown resume to bootstrap data/candidate-profile.yaml and data/star-stories.md on first setup.

4. How It Works

The workflow connects verified candidate information with a target job description to produce ready-to-send documents:

[Candidate Profile (YAML)] + [Job Description (URL / Text)]
                          |
                          v
         [Pick Me: Profile & JD Skill Matching]
                          |
        [Interactive Draft Review & Approval Plan]
                          |
                          v
  [Typst CLI Engine] ---> [PDF in Documents/pick-me-output/] + [CSV Log]
  1. Step 1: Input & Ingestion: You provide a target job posting URL or pasted text. Pick Me loads your verified background from data/candidate-profile.yaml as the sole source of truth.
  2. Step 2: Analysis & Approval: The skill analyzes role requirements, drafts tailored resume bullet points and cover letter paragraphs, and presents them in an interactive plan for your review and approval.
  3. Step 3: Compilation & Output: Upon approval, Pick Me populates the Typst template, invokes the Typst compiler via PowerShell, and saves the final PDF into ~/Documents/pick-me-output/. It then logs the submission details into data/applications.csv.

5. The Tech Stack

  • Agent Environment: Google Antigravity — Manages the plugin runtime, providing the interface for inline skills and the isolated crawler subagent.
  • Document Typesetting: Typst — A modern markup-based typesetting system chosen over LaTeX because it compiles to vector PDF in milliseconds and produces clean, parser-friendly single-column documents without needing heavy toolchains.
  • Automation & Scripting: PowerShell — Handles directory management, environment checks, and compilation scripts on Windows.
  • Data Grounding & Tracking: YAML & CSV — Plain text files (candidate-profile.yaml and applications.csv) keep all data human-readable, easily editable in any text editor, and version-controlled without running a background database server.
  • Web Verification: Playwright & read_url_content — Verifies live HTTP status and parses job descriptions from modern JavaScript-heavy career boards.
Quokka AIQuokka