vibe-resume CLAUDE.md

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Development commands

uv venv && uv pip install -e ".[dev]"   # one-time setup (installs `vibe-resume` entry point)

uv run vibe-resume <subcommand>         # canonical invocation (post-install)
uv run python -m vibe_resume <subcommand>  # equivalent module invocation

uv run pytest tests/                    # full suite (currently 50 tests)
uv run pytest tests/test_review.py      # single file
uv run pytest tests/test_extractors.py::test_cursor_happy_path   # single test

uv run ruff check .                     # lint
uv run ruff check --fix .               # autofix import order, trivial issues

No separate build step — the project is a pure-Python CLI. pandoc is an optional system dep (PDF rendering only); claude -p is an optional binary (enricher LLM path; falls back to rule-based summaries when absent).

Pipeline architecture

The CLI orchestrates a six-stage pipeline. Each stage writes to data/cache/ or data/resume_history/ so later stages can re-run independently:

extract → aggregate → enrich → render → review → trend

• extract (src/vibe_resume/core/runner.py::run_extractors) — drives all extractors via ThreadPoolExecutor(max_workers=4) with a rich.progress display. Each extractor returns list[Activity] (schema in src/vibe_resume/core/schema.py) which is serialized to data/cache/<name>.json. Failure of one extractor is isolated and does not abort the batch. git_repos has a 120s wall-clock deadline inside _find_repos to survive FUSE/broken-symlink stalls.

• aggregate (src/vibe_resume/core/aggregator.py) — groups activities by project and writes data/cache/_project_groups.json + _window_stats.json. Project names are abstracted when config.privacy.abstract_tech: true.

• enrich (src/vibe_resume/core/enricher.py) — feeds each group through a locale-shaped LLM prompt (XYZ verbs for en_*, noun-phrase for zh_*/ja_JP/ko_KR/ de_DE/fr_FR). --tailor <JD.txt> injects extracted keywords so achievements bias toward the JD when the raw activity supports it (never inventing matches). Output: per-group bullets, role_label, summary written back into _project_groups.json.