llm-job-pipeline: Command for Claude Code

/jobs-review

One entry point for reviewing vacancies after scoring. The first argument picks the mode:

Invocation	Mode
/jobs-review (no arg)	Status menu — print counts, let the user pick
/jobs-review apply	Deep structured interview of liked vacancies
/jobs-review archive	Interactive archival of low-scoring unseen vacancies
/jobs-review vac [list|show|mark|open|companies] ...	Thin triage CLI (pass-through to scripts/vac.py)

Never auto-drop into apply. With no argument, show the menu and wait.

Reply in the user's product language. Read ## OUTPUT_LANGUAGE from config/user_profile.md (resolve with python3 -c "import sys;sys.path.insert(0,'scripts');import product_language as p;print(p.resolve())" → en/ru) and write ALL of your chat — menus, prompts, interview questions, summaries — in that language.

After any mode that mutates state (apply, archive) — run the Publish step at the end (see bottom). vac mutations the user makes by hand can also publish on request, but the menu/CLI itself does not auto-publish.

---

Mode: status menu (no argument)

Print the current review backlog so the user knows what is worth doing, then ask which mode to enter.

from config import LLM_SCORE_THRESHOLD
from database_supabase import load_vacancies, get_protected_ids

v = load_vacancies()
threshold = LLM_SCORE_THRESHOLD  # default = 20

liked = [x for x in v.values() if x.get('status') == 'liked']

protected_ids = get_protected_ids()
archivable = [
    (vid, vac) for vid, vac in v.items()
    if vid not in protected_ids
    and vac.get('llm_score') is not None
    and vac['llm_score'] < threshold
]

Render:

JOBS REVIEW
============================================================
  {len(liked):3d} liked awaiting apply-decision   → /jobs-review apply
  {len(archivable):3d} low-score archivable (< {threshold})    → /jobs-review archive
  {len(v):3d} vacancies in DB total              → /jobs-review vac list
============================================================