give-me-job

give-me-job

give-me-job is an AI agent toolkit for Korean job applications. It helps you turn scattered career notes, a Korean job post, and company context into a focused application package: JD analysis, evidence-grounded 자기소개서 drafts, HR risk review, interview defense, and a manual submission checklist.

Most job-application tools optimize for speed. give-me-job optimizes for confidence: every strong claim should trace back to your resume.md, every risky sentence gets reviewed before final use, and every final answer should be something you can defend in an interview.

Use it when you want an agent to help with 취업준비 without inventing experience, exaggerating metrics, or auto-submitting applications. The current demo focuses on a backend role, but the workflow is designed for Korean job applications across tech, business, support, creative, and operations roles. This repository prepares the materials; you stay in control of review and final submission.

Korean search keywords: 자기소개서, 취업준비, 한국 취업, 취업 에이전트, 개발자 취업준비, 개발자 자기소개서.

Use agent.md as the orchestrator for the full workflow. Repository-level agent instructions are in AGENTS.md. Korean documentation is available at docs/README-ko.md.

See the output before you install — a complete demo application package is committed to this repository: examples/demo-new-grad-backend/applications/demo-cloud-backend/

Is This For You?

A good fit

• You are applying to Korean companies and already use a coding agent (Claude Code, Codex, or OpenCode).
• You have a resume or portfolio, and the hard part is turning it into 자기소개서 sentences you can defend.
• You want the cover letter reviewed before submission, not written faster.

Not a fit yet

• You cannot run local Node.js tools. The CLI, TUI, and local dashboard require Node.js 18.17+; AI drafting still runs in your existing coding agent.
• You want automated mass applying. This toolkit refuses to submit, log in, send email, or bypass CAPTCHA by design.
• You want the agent to fill gaps in your experience. It stops and asks instead of inventing.

Features

• Resume evidence intake: structure user-provided career facts into reusable resume.md evidence.
• JD analysis: extract role requirements, evaluation criteria, keywords, gaps, and risks from Korean job posts.
• Company-values analysis: use optional mission, culture, values, or talent-profile material for positioning.
• Korean cover-letter drafting: write answers grounded in resume.md, JD analysis, and optional company context.
• Career-type playbooks: new graduates and experienced hires are drafted and reviewed against different criteria, including 이직 사유 and short-tenure handling for experienced candidates.
• HR review: check exaggeration, unsupported claims, company-name residue, wrong sub-role targeting, and pre-submission blockers.
• Interview preparation: generate follow-up questions and evidence-backed answer points for interview defense.
• Application packaging: create one applications/<company-role>/ package with a manual checklist.
• Job operations: paste multiple posting URLs (one per line), import CSV or JD text, and keep active lists clean by removing jobs after their Seoul-time deadline.
• Resumable validation: separate template structure checks from readiness checks tied to resume, JD, question, answer, evidence, final-text, and HR-review hashes.
• Recruitment tracking: record user-confirmed submissions, arbitrary stages, Seoul-time schedules, interview debriefs, and confirmed outcomes.
• Shared surfaces: operate the same local data and validation through the agent, CLI, keyboard TUI, or a 127.0.0.1 web dashboard.
• Job-source tools: normalize a user-supplied public HTTPS job page, with dedicated extraction for JobKorea, Linkareer, SK Careers, and LG Careers. Automated discovery remains outside the product scope.
• No credentials: nothing here needs an API key, access token, or approved API access.

Documentation

Document	Purpose
Korean README	Korean detailed guide and documentation hub
Quickstart	Minimal local workflow for one application package
npm Install	npm-based skill and agent installation
Platform Support	Windows PowerShell, Ubuntu/Linux, and macOS support
Safety Policy	Allowed actions, disallowed actions, and blocker examples
Release Checklist	Pre-release validation checklist
Competitive v1 Roadmap	Korea-only product roadmap
Job Source Integrations	Manual URL intake support and automated discovery status
Integrated Workflow	Jobs, readiness, tracking, TUI, dashboard, and agent-request commands
Agent Evaluation	Fictional new-grad, experienced, and non-technical verification scenarios

Requirements

Use Node.js 18.17 or newer.

node --version
npm --version

The local tools support:

• Windows PowerShell
• Ubuntu/Linux shell
• macOS shell

To use this repository directly:

git clone https://github.com/kyoungbinkim/give-me-job.git
cd give-me-job
npm test

Installation

The simplest install path is npx:

npx give-me-job@latest install

Keep the npx give-me-job@latest prefix for later commands. If you want the shorter give-me-job command, install the CLI globally as shown below.

This installs the eight domain skills plus the give-me-job orchestrator agent for supported coding agents in your user profile. The support bundle includes agent.md, tools/, templates/, and validation fixtures. Claude Code also gets tool-shaped skills such as give-me-job-fetch-jobs; OpenCode also gets custom tool definitions under its tools directory.

You can also install the CLI globally:

npm i -g give-me-job
give-me-job install

Claude Code users can add the domain skills as a plugin instead, without npm:

/plugin marketplace add kyoungbinkim/give-me-job
/plugin install give-me-job

The plugin route installs the domain skills only. The npm installer additionally writes the give-me-job orchestrator agent, the tool-shaped workflow skills, and the support bundle (agent.md, tools/, templates/, fixtures), which the full workflow needs. Use npx give-me-job@latest install for the complete setup.

Without --target, the installer asks which AI agent to install to. Use the arrow keys to choose codex, claude, opencode, or all.

Install every supported target:

give-me-job install --target all

Install one target:

give-me-job install --target codex
give-me-job install --target opencode
give-me-job install --target claude-code

Install into the current project instead of your user profile:

give-me-job install --scope project --target all

Preview changes before writing files:

give-me-job install --dry-run

By default, the installer refuses to overwrite files that differ from the packaged version. Use --force only when you intentionally want to replace existing installed files. The installer writes timestamped .bak-* backups first.

give-me-job install --target codex --force

Check the installation:

give-me-job doctor

Uninstall files tracked in the install manifest:

give-me-job uninstall --target all

uninstall removes only files recorded in ~/.give-me-job/install-manifest.json and only when the current file still matches the installed hash.

Install Paths

These paths match the official discovery locations for Codex, Claude Code, and OpenCode skills and agents.

Default user-scope install paths:

Codex:       ~/.agents/skills/<skill>/SKILL.md
Codex:       ~/.codex/agents/give-me-job.toml
Codex:       ~/.codex/give-me-job/

OpenCode:    ~/.config/opencode/skills/<domain-skill>/SKILL.md
OpenCode:    ~/.config/opencode/agents/give-me-job.md
OpenCode:    ~/.config/opencode/tools/give_me_job_<tool>.js
OpenCode:    ~/.config/opencode/give-me-job/

Claude Code: ~/.claude/skills/<domain-skill>/SKILL.md
Claude Code: ~/.claude/skills/give-me-job-<tool>/SKILL.md
Claude Code: ~/.claude/agents/give-me-job.md
Claude Code: ~/.claude/give-me-job/

Project-scope install paths:

Codex:       .agents/skills/<skill>/SKILL.md
Codex:       .codex/agents/give-me-job.toml

OpenCode:    .opencode/skills/<domain-skill>/SKILL.md
OpenCode:    .opencode/agents/give-me-job.md
OpenCode:    .opencode/tools/give_me_job_<tool>.js

Claude Code: .claude/skills/<domain-skill>/SKILL.md
Claude Code: .claude/skills/give-me-job-<tool>/SKILL.md
Claude Code: .claude/agents/give-me-job.md

Quickstart

Create a workspace and add your factual resume:

mkdir my-job-search
cd my-job-search
# Create resume.md here, or copy your existing resume.md into this folder.

Save one supplied JD as job-import.json:

{
  "kind": "text",
  "input": "Paste the complete job description here.",
  "metadata": {
    "company": "Kakao",
    "title": "Backend Developer",
    "role": "backend"
  }
}

Import it and copy the returned jobId:

npx give-me-job@latest jobs import --workspace . --input job-import.json
npx give-me-job@latest jobs list --workspace . --format json

Prepare the linked package and generate the agent prompt:

npx give-me-job@latest application prepare --workspace . --job-id "JOB_ID_FROM_PREVIOUS_COMMAND" --role backend
npx give-me-job@latest request --workspace . --task assess --job-id "JOB_ID_FROM_PREVIOUS_COMMAND"

Run that prompt in Codex, Claude Code, or OpenCode. After the agent writes and reviews the package, run readiness validation and open the local dashboard:

npx give-me-job@latest application validate --workspace . --package-path "PACKAGE_PATH_FROM_PREVIOUS_COMMAND" --mode ready
npx give-me-job@latest dashboard --workspace .

The dashboard opens only on 127.0.0.1; stop it with Ctrl+C. For CSV and URL imports, tracking, TUI use, and exact validation states, see Quickstart and Integrated Workflow.

Workflow

JD URL / JD text
        |
        v
agent.md intake ---- missing or weak ----> resume-intake
        |                                      |
        v                                      v
existing resume.md <---------------------- resume.md
        |
        v
jd-analyzer
        |
        v
JD analysis <---- optional ---- company-values-analyzer
        |
        v
cover-letter-writer
        |
        v
hr-reviewer
        |
        v
interview-prep
        |
        v
application-packager
        |
        v
applications/<company-role>/
        |
        v
User reviews and submits manually

The orchestrator must stop when required inputs are missing, when resume.md does not support a strong claim, or when the next action would submit, send, log in, bypass CAPTCHA, or transmit personal information.

Package Output