resume-convert

name

resume-convert

description

Convert resumes between PDF, Markdown, and the canonical JSON Resume format (ResumeJson). Use this whenever the user wants to extract a resume from a PDF into structured JSON, turn a Markdown resume into JSON, or render a resume JSON back to Markdown. Triggers on: "extract resume from PDF", "parse my resume", "resume to JSON", "convert resume to markdown", "resume from markdown", or any reference to ResumeJson / JSON Resume. All AI steps run on the current Claude session — no OpenAI, no API key.

resume-convert

Convert resumes between PDF / Markdown / JSON. The canonical format is ResumeJson (JSON Resume). The AI steps are performed by you, Claude, in this session — there is no external model and no API key. A small TypeScript toolbelt CLI handles the deterministic, non-AI work (rendering, schema validation).

The toolbelt CLI

Deterministic operations run through the bundled CLI. In this repo (development), invoke it from the repo root as:

npm run resumejson_claude -- <command> [args]

Once the package is installed, npx resumejson_claude <command> works the same.

Command	Purpose
schema resume	Print the JSON Schema for ResumeJson — the exact contract to produce
validate -i <file> --schema resume	Zod-validate a JSON file; exits non-zero and lists problems on failure
to_markdown -i <resume.json> -o <out.md|->	Render ResumeJson → Markdown (deterministic)
to_pdf -i <resume.json> -o <out.pdf>	Render ResumeJson → PDF (deterministic; Mustache + headless Chrome)

The contract: ResumeJson

Every step reads or writes a ResumeJson document. Never guess the shape — fetch it once at the start of a task:

npm run resumejson_claude -- schema resume

The canonical shape carries all section keys (basics, work, education, skills, …); a section the person doesn't have is null or [], never absent.

Capabilities

from_pdf — PDF → ResumeJson (AI, you)

1. Read the PDF with the Read tool (it renders pages as images; use your vision).
2. Fetch the contract: schema resume.
3. Produce a ResumeJson object with every key present (value, null, or []). Do not invent facts — use null for anything the PDF doesn't state.
4. Write it to <name>.resume.json.
5. Validate + self-correct (see below).

from_markdown — Markdown → ResumeJson (AI, you)

Same as from_pdf, but read the Markdown resume as text and structure it into ResumeJson. Preserve wording; only reshape into the schema.

to_markdown — ResumeJson → Markdown (deterministic, CLI)

No AI. Just run:

npm run resumejson_claude -- to_markdown -i <name>.resume.json -o <name>.resume.md

(Use -o - to stream to stdout.) It validates the input first, so a malformed resume fails loudly rather than rendering garbage.

to_pdf — ResumeJson → PDF (deterministic, CLI)

No AI. Renders the resume through a Mustache HTML template and prints it with headless Chrome:

npm run resumejson_claude -- to_pdf -i <name>.resume.json -o <name>.resume.pdf

Pass -t <template.mustache.html> to use a custom template. Requires a headless Chrome (installed with Puppeteer); the default template pulls Bootstrap from a CDN, so PDF rendering needs network access.

Validate + self-correct (required after every AI step)

This is how we get schema-guaranteed output without a structured-output API:

1. Write your JSON to a file.
2. Run validate -i <file> --schema resume.
3. If it exits non-zero, read each reported problem (they are path: message), fix the JSON, and re-validate.
4. Repeat until it passes. Only then present a short human summary of what you extracted (name, counts of work/education/skills entries, anything you left null).

Notes

• This skill is deliberately narrow (conversion only). ATS scoring / review / optimization live in the sibling resume-ats skill.
• No credentials are ever needed — the reasoning is you, the deterministic work is local TypeScript.