/generate-resume

/generate-resume

Generate a Harvard OCS-style LaTeX resume from the user's Obsidian notes.

Plugin root: ${CLAUDE_PLUGIN_ROOT} Skill source of truth: ${CLAUDE_PLUGIN_ROOT}/skills/obsidian-resume/

Follow these steps in order. Do not skip steps.

---

Step 1 — Locate the input

The user may have passed input as $ARGUMENTS. Resolve it like this:

1. If $ARGUMENTS is a directory → it's an Obsidian vault. Run the export script on it (see Step 1a below). This is the recommended path.
2. If $ARGUMENTS is a .json file → read it with the Read tool. This is a pre-exported obsidian_export.json.
3. If $ARGUMENTS is a .md file → read it. Treat as raw resume notes.
4. If $ARGUMENTS is empty → ask the user:

   "Give me a path to your Obsidian vault folder, a pre-exported obsidian_export.json, or paste markdown notes here."

Use Bash test -d "$ARGUMENTS" to distinguish a directory from a file.

Do not proceed until you have actual content.

---

Step 1a — Run the export script (when input is a folder)

When the user passes a vault path, invoke the bundled Python exporter via the Bash tool — do not reimplement the parsing inline.

OUT="$(mktemp -t obsidian_export.XXXXXX.json)"
python3 "${CLAUDE_PLUGIN_ROOT}/skills/obsidian-resume/scripts/export_vault.py" \
  --vault "$ARGUMENTS" \
  --out "$OUT"
echo "$OUT"

Then read the JSON file at the path printed by echo. The script handles:

• Walking the vault and skipping .obsidian/, .trash/, and other dot-directories
• Filtering for career-tagged notes (#career, #resume, #job, #experience, etc.) or career-named folders (Career, Job, Resume, CV, Work, Experience, Portfolio, ...)
• Parsing YAML frontmatter, inline tags, and Dataview key:: value fields
• Producing the JSON schema documented in references/export_schema.md

If the script fails:

• python3: command not found → tell the user they need Python 3 installed and stop.
• Vault not found → the path was wrong; ask the user for the correct path.
• Anything else → show the user the error and stop.

If the script returns zero notes:

Re-run with --all to include every .md file, or ask the user which folder/tag to scope to:

python3 "${CLAUDE_PLUGIN_ROOT}/skills/obsidian-resume/scripts/export_vault.py" \
  --vault "$ARGUMENTS" --all --out "$OUT"

Once you have the JSON, read it with the Read tool and proceed to Step 3.

---

Step 2 — Read the references

Before generating LaTeX, read these in full:

1. ${CLAUDE_PLUGIN_ROOT}/skills/obsidian-resume/SKILL.md — full workflow
2. ${CLAUDE_PLUGIN_ROOT}/skills/obsidian-resume/references/harvard_template.md — LaTeX preamble + section examples
3. ${CLAUDE_PLUGIN_ROOT}/skills/obsidian-resume/references/export_schema.md — JSON schema (only if working from JSON)

These files are the authoritative spec. The rules in them override anything else.

---

Step 3 — Parse the input into a profile

Build the structured profile object defined in SKILL.md Step 1.

Rules:

• Frontmatter keys (title:, company:, dates:, tags:) map directly to profile fields
• ## Experience, ## Skills headings signal sections
• Bullets under a job heading become bullets[]
• Dataview fields (company:: Acme) are treated like frontmatter
• Tags #career, #resume, #job mark notes as relevant
• If a field is missing, leave it out. Do NOT invent data.
• Preserve the user's date format — do not reformat

---

Step 4 — Generate the LaTeX resume

Apply every rule from references/harvard_template.md:

• Use the full preamble exactly as written there
• Section order: Education → Experience → Projects → Skills → Awards
• ALL CAPS section headers with \hrule
• Dates right-aligned with \hfill
• Bullets: past-tense action verbs, quantified impact where the notes support it
• Escape LaTeX specials: & % $ # _ { } ~ ^ \
• Include \usepackage[utf8]{inputenc} and \usepackage[T1]{fontenc} for accented names
• 1 page target (2 max for 10+ years experience)

Run the output checklist from SKILL.md:

• All special characters escaped
• inputenc and fontenc included
• Every \begin{} matches an \end{}
• Dates right-aligned with \hfill
• Section headers use \section*{} + \hrule
• No invented data
• File would compile with pdflatex (mentally trace)

---

Step 5 — Save the output

Ask the user where to save:

"Where should I save the .tex file? (default: ./resume.tex)"

Write the complete .tex to that path with the Write tool. Then print compile instructions:

pdflatex resume.tex
pdflatex resume.tex   # run twice for correct spacing

If pdflatex isn't installed, point to:

• macOS: brew install --cask mactex
• Ubuntu: sudo apt install texlive-full
• Windows: MiKTeX from miktex.org

---

Step 6 — Offer the next step

After saving, offer:

"Resume saved. Want me to find live job listings that match this profile? Run /find-jobs with the same input."

---

Important

• Never invent achievements, metrics, dates, employers, or education. Only use what's in the notes.
• If the notes are sparse, write strong qualitative bullets — do not pad with fake metrics.
• Vietnamese / accented characters work natively with inputenc + fontenc. Do not transliterate names.