Skip to content
Kevin拾玖秋— home
中文
All projects

05Open Source · AI Coding Agent Skill

job-application-assistant

Open Source AI Skill for Claude & Codex

Three views of job-application-assistant: overview, application dashboard and company background report (sample data)
Role
Author and maintainer.
Stack
Open Source · Claude · Codex · Agent Skill
Links
GitHub(opens in a new tab)

01

Problem

01.1Overview

An open-source skill for Claude and Codex that helps individual job seekers find roles, research companies, fill forms carefully and track every application.

01.2Context

A skill packages a workflow so a coding agent like Claude or Codex can run it reliably.

This one is for individual job seekers — not a recruiting SaaS. No server, no accounts, no cloud sync.

01.3Problem

Job seekers repeat the same manual work: copying job details, checking whether a company is trustworthy, and tracking applications in a spreadsheet.

01.4My Role

Author and maintainer.

02

Build

02.1Solution

A skill that guides the agent through a fixed flow: find roles → confirm target companies → research the company → decide → assist with the form → the user confirms submission → track and follow up.

Application records are stored in a local JSON file. The dashboard and company reports are self-contained static HTML regenerated from that data.

Features

  • 01Job Application Dashboard
  • 02Job details side panel
  • 03Company due-diligence report
  • 04Careful form assistance for common ATS forms
  • 05Duplicate-application check
  • 06Local data, no accounts

02.2Architecture

  1. SKILL.md
  2. references/
  3. scripts/
  4. Local JSON
  5. Static HTML dashboard
fig. 05a — skill structure
05One application, end to end

  1. Find Roles

    Find hiring companies and their career pages

  2. Research

    Company research — public sources only

  3. Fill Form

    Assist with the form in a visible browser

  4. User Confirms

    Waiting for the user to submit — never automatic

  5. Track

    Track the application and follow up

One run of the pipeline, step by step.

02.3Key Decisions

  1. 01

    Human confirmation by design

    The skill never submits an application on its own. Every real submission needs explicit confirmation.

  2. 02

    No fabrication

    It does not invent resume experience, and due-diligence conclusions must have public sources.

  3. 03

    Stops at CAPTCHAs

    When a CAPTCHA appears, control goes back to the user.

03

Engineering

03.1Engineering

  • Packaged as a skill: SKILL.md, scripts/ and references/.
  • Tracking and report scripts use only the Python standard library.
  • Form assistance runs in a visible browser via Playwright.

04

Notes

04.1What I Learned

A good agent skill is mostly about where the agent must stop.

04.2Tech Stack

  • Claude
  • Codex
  • Agent Skill
  • Python
  • Playwright
  • Local-first