How to Use the /apply Command with a Job URL in ai-job-search

The /apply command accepts a job posting URL or raw text and executes a six-step, two-agent workflow that fetches the posting, evaluates your fit against local profile data, drafts tailored LaTeX documents, critiques them through a dedicated reviewer agent, and compiles ATS-ready PDFs.

The ai-job-search repository by MadsLorentzen provides an automated job application pipeline that transforms raw postings into polished application materials. At the center of this system is the /apply slash command, defined in .claude/commands/apply.md, which orchestrates a sophisticated multi-agent process to generate customized CVs and cover letters from a single URL or text block.

Architecture and Workflow

The /apply command implements a two-agent architecture comprising a DRAFTER and a REVIEWER. This design keeps processing token-efficient by passing data inline between steps rather than repeatedly reading from disk, ensuring that once a file is loaded it remains in memory for subsequent operations.

Step 0 — Parse Input and Fetch URL

The command first detects whether you provided a URL or plain text. If a URL is supplied, the system invokes WebFetch to retrieve the raw HTML of the job posting. It then extracts essential metadata including company name, role title, department, location, and language settings to configure the downstream workflow.

Step 1 — Evaluate Fit (DRAFTER)

The DRAFTER agent reads the evaluation framework from .claude/skills/job-application-assistant/04-job-evaluation.md and your candidate profile from .claude/skills/job-application-assistant/01-candidate-profile.md. It optionally executes salary_lookup.py to inject market-salary benchmarks. The agent produces a structured fit assessment including skills match, experience alignment, culture fit, and salary expectations, then prompts you to confirm before proceeding.

Step 2 — Draft CV and Cover Letter (DRAFTER)

Upon confirmation, the DRAFTER reads style guides and LaTeX templates—specifically cv/main_example.tex and cover_letters/cover.cls—to generate cv/main_<company>_<role>.tex and cover_letters/cover_<company>_<role>.tex. Every requirement from the job posting is mapped to specific sections in the documents, respecting the language settings and auditing grounding against your master profile.

Step 3 — Research and Critique (REVIEWER)

A separate REVIEWER agent receives the drafts inline (without additional file reads) and performs research, factual grounding checks, and style analysis. It returns a structured critique divided into two parts: Part A, a JSON edit list for automated changes, and Part B, narrative suggestions requiring manual incorporation.

Step 4 — Revise Based on Feedback (DRAFTER)

The DRAFTER automatically applies the JSON edits from Part A and manually integrates the narrative suggestions from Part B. This step guarantees no fabrications are introduced and ensures the drafts remain consistent with your established candidate profile.

Step 5 — Compile and Inspect PDFs (DRAFTER)

The command runs lualatex for the CV and xelatex for the cover letter, generating final PDFs. It verifies page count, checks for orphaned sections, and confirms ATS-readability of the text layer. If issues are detected, the DRAFTER re-edits the LaTeX source and recompiles until all visual and technical checks pass.

Step 6 — Present Final Output

Finally, the command executes the master verification checklist defined in CLAUDE.md, displaying a concise summary of tailoring decisions and listing the created files with their paths.

Command Syntax and Examples

You can invoke the /apply command from any interface that respects the slash-command syntax. The following patterns are supported:

# Standard usage with a direct job URL

/apply https://jobs.example.com/software-engineer-12345

# Using raw posting text for offline or paywalled listings

/apply <<EOF
We are looking for a Senior Data Scientist with 5+ years of experience...
EOF

During execution, you will see interactive prompts. After Step 1, you receive an evaluation report similar to:

Skills match: 7 / 9
Experience match: strong
Salary benchmark: $120k-$140k (Denmark)
Overall fit: strong
Proceed with drafting? (yes/no)

Once drafting completes, the reviewer returns structured feedback. Example JSON edit list (Part A):

[
  {
    "file": "cv/main_acme_corp_senior_data_scientist.tex",
    "old_string": "Machine Learning Engineer",
    "new_string": "Senior Machine Learning Engineer",
    "reason": "keyword match (senior)"
  }
]

After revision and compilation, the terminal displays:

✅ All checks passed.
📄 Created files:
  - cv/main_acme_corp_senior_data_scientist.tex
  - cover_letters/cover_acme_corp_senior_data_scientist.tex

Key Implementation Files

• .claude/commands/apply.md — The main command definition containing the six-step workflow logic and agent instructions.
• .claude/skills/job-application-assistant/04-job-evaluation.md — Evaluation framework used to score job fit during Step 1.
• .claude/skills/job-application-assistant/01-candidate-profile.md — Source-of-truth candidate profile referenced throughout drafting.
• salary_lookup.py — Optional Python tool invoked during fit evaluation to retrieve market salary data.
• cv/main_example.tex — Master LaTeX CV template providing structural reference for generated documents.
• cover_letters/cover.cls — LaTeX class file defining cover letter styling and layout.
• CLAUDE.md — Global configuration and verification checklist ensuring factual accuracy and ATS compatibility.

Summary

• The /apply command in ai-job-search automates the full job application pipeline from URL to PDF.
• It uses a two-agent system (DRAFTER and REVIEWER) with inline data passing to optimize token usage.
• The workflow covers six distinct phases: input parsing, fit evaluation, document drafting, content critique, revision, and PDF compilation.
• Key files include .claude/commands/apply.md for logic and CLAUDE.md for verification standards.
• Supports both URL fetching via WebFetch and direct text input for maximum flexibility.

Frequently Asked Questions

Can I use the /apply command with plain text instead of a URL?

Yes. While the command automatically detects and fetches URLs using WebFetch, you can paste raw job description text directly after the /apply command. This is useful for paywalled job boards or when you have the description in a local file.

What LaTeX engines does the command use for compilation?

According to the source code in .claude/commands/apply.md, the DRAFTER uses lualatex to compile the CV and xelatex to compile the cover letter. This dual-engine approach ensures proper font rendering and layout consistency across both documents.

How does the reviewer agent provide feedback?

The REVIEWER agent returns feedback in two parts: Part A is a structured JSON array specifying exact string replacements for automated application, and Part B contains narrative suggestions that require manual review. The DRAFTER then applies Part A automatically while carefully integrating Part B.

Is the salary lookup automatic when using /apply?

The salary lookup via salary_lookup.py is optional. During Step 1 (Evaluate Fit), the DRAFTER may invoke this tool to add market-salary context to the evaluation report, but the workflow proceeds regardless of whether salary data is available.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}