cover-letter-builder

Cover Letter Builder

Inputs

Field	Source
Job title	User message or job posting
Company name	User message or job posting
Job description	Pasted text, or file path to a .pdf / .txt
CV / resume	File path (.pdf, .md, .tex, .txt) or pasted text — asked in Step 0 if not provided
Application channel	Default: standard; options: linkedin
Repo path	Default: current working directory (.)

If any required field is missing and cannot be inferred, ask once — all at once.

---

Step 0: Intake

Run this before scanning the repo.

Has the user already provided a CV, resume, or background text?
  YES → proceed to Step 1 with that as the primary source
  NO  → ask the user (single message):

"Do you have an existing CV or resume? If yes, provide the file path or paste the text here. If not, I'll scan the current repo for your background."

Also confirm the job description is complete. If the user only provided a job title and company name, ask for the full JD now.

If the user has no CV and the repo scan (Step 1) finds nothing useful: switch to inline intake — ask these questions in one message:

• Current role and company
• 2-3 most relevant past roles (title, company, one key outcome each)
• Top 5-6 skills most relevant to the target role
• Education (brief)

Use the answers as the profile source.

---

Step 1: Scan Sources

If the user provided a CV file (Step 0)

Use it as the primary source. Extract text based on format:

PDF:

# Best layout preservation (poppler-utils)
pdftotext -layout cv.pdf -

Fallback chain:

1. pdftotext -layout file.pdf -
2. pdftotext file.pdf -
3. If pdftotext is unavailable: tell the user and ask them to paste the CV text

Markdown / plain text: read directly — no transformation needed.

LaTeX: strip commands, extract readable text:

sed 's/\\[a-zA-Z]*{[^}]*}//g; s/\\[a-zA-Z]*//g; s/[{}]//g'

Or pandoc -t plain file.tex if available.