Understanding the Role of CLAUDE.md in the AI Job Search Framework Architecture

CLAUDE.md serves as the central candidate profile and single source of truth for the AI Job Search framework, storing all personal, educational, and professional data that AI agents consume to evaluate job fit, tailor application documents, and verify factual accuracy.

The MadsLorentzen/ai-job-search repository implements a thin-pointer design that treats CLAUDE.md as the authoritative datastore for the entire job search pipeline. Instead of scattering candidate information across multiple configuration files, the framework centralizes identity details, work history, skills, and behavioural data in this single markdown file, enabling consistent, drift-free automation across Claude Code, Google Antigravity, Codex, and other AI agents.

What Is CLAUDE.md in the AI Job Search Framework?

CLAUDE.md is the canonical candidate profile document located at the repository root. It functions as the master record containing comprehensive applicant information:

• Identity data: Name, location, contact details
• Educational background: Degrees, certifications, institutions
• Professional experience: Work history, achievements, responsibilities
• Technical and soft skills: Competency matrices and proficiency levels
• Behavioural attributes: Communication style, work preferences, cultural fit indicators

According to the repository structure defined in CLAUDE.md itself, this file is the single source of truth that prevents data duplication across the framework’s various skill definitions and command scripts.

Thin-Pointer Architecture: Single Source of Truth

The framework adopts a thin-pointer design pattern explicitly documented in AGENTS.md. This architectural choice ensures that:

• No data drift occurs: AI agents read candidate data directly from CLAUDE.md rather than maintaining cached copies or separate databases
• Updates propagate instantly: Modifying the profile immediately affects all downstream workflows without requiring synchronization across multiple files
• Consistency is guaranteed: CV generators, cover letter tools, and job fit evaluators all reference identical candidate information

As implemented in MadsLorentzen/ai-job-search, this design eliminates fragmentation where resume builders might otherwise use outdated skill lists or location data.

Workflow Integration Across the Framework

The canonical workflow specifications located under the .claude/ directory reference CLAUDE.md to drive three core operational phases:

Job Fit Evaluation

Skill definitions in .claude/skills/job-application-assistant/ (such as the 01-*.md files) parse the candidate profile to match qualifications against job descriptions. The framework evaluates alignment between the applicant's documented competencies and role requirements by referencing the thin-pointer design principles outlined in AGENTS.md.

Document Generation

CV and cover letter generation tools extract specific candidate attributes—such as location, name, and key achievements—from CLAUDE.md to personalize output. CLI tools located in .agents/skills/ invoke these workflows, relying on the profile data to inject accurate, context-aware content into application materials.

Verification and Quality Assurance

Before generating final deliverables, the framework executes a verification checklist against CLAUDE.md to confirm factual accuracy. This ensures that claimed skills, employment dates, and educational credentials match the canonical profile, preventing hallucinations or inconsistencies in AI-generated documents.

Programmatic Access to Candidate Data

Developers and automation scripts interact with CLAUDE.md using standard file parsing techniques. The markdown structure uses YAML-style sections that facilitate programmatic extraction.

Python Parsing Implementation

The following Python helper demonstrates how to load and parse the candidate profile:

import yaml, pathlib

def load_profile():
    path = pathlib.Path(__file__).parent.parent / "CLAUDE.md"
    # Strip the Markdown front‑matter and parse the remaining YAML‑style sections

    with open(path, "r", encoding="utf-8") as f:
        lines = [l for l in f.readlines() if not l.startswith("#")]
    # Simple parsing: split on double newlines and extract key‑value pairs

    profile = {}
    for block in "\n".join(lines).split("\n\n"):
        if ":" in block:
            key, val = block.split(":", 1)
            profile[key.strip()] = val.strip()
    return profile

candidate = load_profile()
print(candidate["Name"], candidate["Location"])

Bash Extraction for Shell Workflows

Shell scripts can extract specific fields using grep and text processing:

#!/usr/bin/env bash

# Load name and city from CLAUDE.md

NAME=$(grep -i '^- \*\*Name:' CLAUDE.md | cut -d':' -f2 | xargs)
CITY=$(grep -i '^- \*\*Location:' CLAUDE.md | cut -d',' -f1 | xargs)

echo "Generating CV for $NAME based in $CITY..."

# Call the CV generation tool with these variables

./tools/generate_cv.sh --name "$NAME" --city "$CITY"

Summary

• CLAUDE.md acts as the canonical candidate profile repository in the MadsLorentzen/ai-job-search framework, consolidating all applicant data in a single, version-controlled location.
• The thin-pointer design documented in AGENTS.md mandates that all AI agents and workflow tools read from this file exclusively, eliminating data synchronization issues.
• Workflow integration spans job fit evaluation, personalized document generation, and automated fact-checking across the .claude/skills/job-application-assistant/ directory.
• Programmatic access is supported through both Python parsing libraries and standard Unix text-processing tools, enabling flexible automation.

Frequently Asked Questions

What specific data should be stored in CLAUDE.md?

CLAUDE.md should contain comprehensive candidate information including full name, location, contact details, educational history, professional experience with quantifiable achievements, technical skills, soft skills, and behavioural attributes. The file uses markdown formatting with YAML-style key-value pairs to facilitate both human readability and programmatic parsing by the framework's AI agents.

How does the thin-pointer design prevent data inconsistencies?

The thin-pointer design prevents inconsistencies by eliminating separate copies of candidate data across the system. Instead of storing name, skills, or location in multiple skill definition files or configuration scripts, all components reference the single CLAUDE.md file. When you update your location or add a new certification, every tool immediately accesses the current version, ensuring CV generators and job fit evaluators never use stale information.

Can I use CLAUDE.md with AI agents other than Claude Code?

Yes. While the filename references Claude, the framework's architecture supports Google Antigravity, Codex, and other AI systems. The markdown-based format with clear section headers allows any agent capable of reading files to parse the profile. The .agents/skills/ directory contains CLI tools that can interface with various AI providers while still sourcing candidate data from the centralized CLAUDE.md profile.

Where are the workflow specifications that use CLAUDE.md located?

The canonical workflow specifications reside in the .claude/skills/job-application-assistant/ directory, typically organized as numbered markdown files (e.g., 01-*.md). These skill definitions explicitly reference CLAUDE.md to drive job-fit evaluation logic and document generation workflows, as detailed in the repository's AGENTS.md documentation.

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