santifer--career-ops/modes/latex-tex.md

Mode: latex-tex — Tailor a user-owned LaTeX CV in place

Opt-in mode for candidates who already maintain a hand-tuned .tex CV. Does not change the global source of truth — cv.md remains the default for evaluations, apply mode, and auto-pipeline. Invoke explicitly via /career-ops latex-tex.

When to use

• User has resume.tex (or config/profile.yml → latex.source) in a supported layout
• User wants JD-tailored bullets/skills while keeping their preamble, macros, colors, and spacing

Supported layouts (v1)

Family	Detection	Editable prose
resumeSubheading	\resumeSubheading + \resumeItem	\resumeItem{...} bullets; \textbf{Category}{: items} skill values
tabularx-itemize	tabularx + itemize, no resume macros	\item body text in the document body

Any other layout → stop with the script error and suggest /career-ops latex (cv.md → career-ops template).

Source file resolution

1. config/profile.yml → latex.source if set
2. Else resume.tex in project root
3. Else cv.tex in project root

If none exist, stop and ask the user to add their .tex file or set latex.source.

# config/profile.yml (optional, user layer)
latex:
  source: resume.tex

Pipeline

1. Resolve source .tex path (see above)
2. Run: node extract-latex-content.mjs <source.tex> --out /tmp/cv-slots-{company}.json
3. If supported: false → show error + hint; do not proceed
4. Read JD (from context, report, or ask user)
5. Tailor only the slots[].text values for JD fit (same ethics as modes/latex.md / pdf):
   • Extract 15–20 JD keywords
   • Reorder bullets by relevance (reorder patch list order if needed; patch ids stay stable)
   • Inject keywords into existing achievements — NEVER invent skills
   • If cv.md exists, cross-check claims against it; omit anything not backed by in-scope sources
6. Write patches file:

{
  "slots": [ "... copy from extract manifest ..." ],
  "patches": [
    { "id": "bullet-0", "text": "Tailored plain-text bullet (no LaTeX escaping — the script escapes)" }
  ]
}

7. Run: node patch-latex-content.mjs <source.tex> /tmp/cv-patches-{company}.json output/cv-{candidate}-{company}-{YYYY-MM-DD}.tex
8. Run: node generate-latex.mjs output/cv-{candidate}-{company}-{YYYY-MM-DD}.tex output/cv-{candidate}-{company}-{YYYY-MM-DD}.pdf --compile-only
9. Report: family, slot count, patched count, .tex path, .pdf path (or compile error)

Requires: tectonic or pdflatex on PATH (same as latex mode).

Ethical rules (mandatory)

Same as modes/latex.md and modes/pdf.md:

• Keywords get reformulated, never fabricated
• Never add tools, skills, or metrics the candidate does not already have in the source .tex or cv.md
• Preserve inline LaTeX markup inside bullets when possible; when rewriting, output plain text in patch JSON (the patch script escapes special characters)
• Do not rewrite preamble, macro definitions, section titles, dates, company names, or job titles unless the user explicitly asks

What this mode does NOT do

• Does not replace cv.md as the system source of truth
• Does not parse arbitrary LaTeX templates
• Does not auto-run during auto-pipeline or evaluation
• Does not submit applications

Relationship to latex mode

Mode	Input	Output
latex	cv.md	career-ops templates/cv-template.tex → .tex + PDF
latex-tex	user's resume.tex	same template shape, tailored prose only → .tex + PDF