Skip to content

0001: One candidate profile, weighted fit factors, CLI-first onboarding

  • Date: 2026-11-09
  • Status: accepted

Problem

A new user could not go from a clean machine to scored jobs without reading the source. Fit scoring read a hand-written watch-profile.md that nothing created. The resume they imported sat unused. The preferences were on/off switches, with no way to say which factor matters more. kairos init stopped at database and API settings.

Options

  1. Keep watch-profile.md as the profile, and have init write it from prompts. This creates a second copy of facts the resume already holds.
  2. Make the base resume the profile, and keep watch-profile.md for optional notes. There is one source for experience, and the notes cover what a CV omits, such as locations or visa status.
  3. Use structured profile fields (years, stack, industries). This is precise, but it duplicates the resume and asks the user for a lot up front.

For weights: boolean switches; free-form numeric weights; or four levels (off/low/medium/high) that map to a points cap.

Choice

Option 2, with four importance levels per factor (skills, industry, career value, work-life balance, foreign companies). Four levels are easy to choose between in a prompt or a select, and the points caps keep the LLM's scoring legible. kairos init walks the whole path: regions and cities, resume, values, scorer backend, harvest, first run. It can also read every answer from --from onboarding.toml, so a clean Docker run can be scripted end to end.