glassdoor-jobs-ai-skills

Imported from yakubmurcek/glassdoor-jobs-ai-skills (AGENTS.md). Install upstream with npx skills add yakubmurcek/glassdoor-jobs-ai-skills. Copyright stays with the author.

Project Context & Agent Instructions

Thesis Context

• Topic: Master's thesis in Economics analyzing Glassdoor job listings in the IT industry in the United States and Czechia.
• Objective: Examine and compare AI skill requirements in U.S. job postings versus Czech job postings.
• Data: Scraped Glassdoor job postings for both the United States and Czechia.
• Main Document: The primary Word document for this thesis is located at docs/Master_thesis.docx.

Project Details

AI skills extraction from Glassdoor job postings for master's thesis (US vs EU). Hybrid extraction strategy: deterministic dictionary matching + OpenAI LLM analysis.

Core Commands

• Python context: Always use uv run python -m ai_skills.cli [subcommand]
• Tests: uv run pytest tests/
• Default input: data/inputs/us_relevant_30.csv
• Be descriptive with output filenames for easy version tracking.

Architecture & Pipeline

• ai_skills/ (core Python package), config/ (TOML settings), data/ (inputs/outputs), analysis/stata/, tests/
• Pipeline Flow: CSV -> Deterministic Extraction -> LLM Analysis -> Merge (Union) -> Stata Formatting -> CSV Output
• Supports resumable processing via checkpoints (run_with_checkpoints()).

Important Rules & Conventions

1. Formatting/Paths: CSV must use ; separator and utf-8-sig encoding. Always use pathlib.Path, never raw strings.
2. Coding Style: Extensive type hints (from __future__ import annotations), descriptive names, lazy initialization where possible, logging everywhere, graceful error handling.
3. Zaokrouhlování čísel v thesis výstupech: Všechny číselné hodnoty (procenta, procentní body, koeficienty, AME, poměry) v textech, tabulkách a interpretacích pro diplomovou práci (docs/, analysis/) vždy zaokrouhlovat na jedno desetinné místo (např. 13,2 %, nikoli 13,19 %). Pokud by zaokrouhlení ztratilo smysl (např. rozdíl 0,04 pp by se stal 0,0 pp), upozornit.

Config Precedence

1. config/settings.local.toml (user overrides, gitignored)
2. config/settings.toml (tracked defaults)
3. Environment variables / .env
4. Hardcoded defaults

• Exception: credentials (OPENAI_API_KEY) MUST come from environment only, never TOML.

Verification Rules (CRITICAL)

When verifying, validating, or checking something:

1. Write a standalone Python verification script (e.g., verify_X.py). Never just eyeball data.
2. Use large random samples (100+ minimum) and deeply check edge cases.
3. Place ALL verification scripts in the verify-scripts/ directory at the repo root. NEVER put them in ai_skills/ or tests/.
4. Command: uv run python verify-scripts/verify_<what>.py