Resume JSON

resumejson_cli Skill

A CLI for converting resumes between PDF / JSON / Markdown and running an AI-powered ATS pipeline (score → questions → answers → review → optimize).

The canonical format is a ResumeJson object validated by a Zod schema. Every command reads or writes that JSON.

Requires: OPENAI_API_KEY environment variable (uses the OpenAI AI SDK, default model gpt-4.1).

Quick Reference

- is accepted as <path> for stdin/stdout.

Setup

cd packages/resumejson_cli
npm install
npm run build

# Set API key
export OPENAI_API_KEY=sk-...

Commands

All commands take -i <input> and -o <output>. AI-backed commands also print a human-readable summary to stdout in addition to writing the JSON output file.

Conversion: from_pdf / to_pdf

# PDF -> resume JSON (uses GPT-4.1 vision over rendered PDF pages, cached in SQLite)
npx resumejson_cli from_pdf -i ./resume.pdf -o ./resume.json

# resume JSON -> PDF (renders via Mustache HTML template + Puppeteer)
npx resumejson_cli to_pdf -i ./resume.json -o ./resume.pdf

Conversion: from_markdown / to_markdown

# Markdown -> resume JSON
npx resumejson_cli from_markdown -i ./resume.md -o ./resume.json

# resume JSON -> Markdown
npx resumejson_cli to_markdown -i ./resume.json -o ./resume.md

ats_score

Quick numerical ATS readiness evaluation — use this as a "before / after" gauge.

npx resumejson_cli ats_score -i ./resume.json -o ./ats_score.json

ats_review

Detailed qualitative ATS review (strengths, weaknesses, suggestions). Feeds ats_optimize.

npx resumejson_cli ats_review -i ./resume.json -o ./ats_review.json

ats_question / ats_answering / ats_answered

Three-step loop that surfaces typical ATS screening questions, answers them from the resume, then folds the answers back in to enrich the resume.

# 1. Generate the questions an ATS would ask about this resume
npx resumejson_cli ats_question  -i ./resume.json                              -o ./questions.unanswered.json

# 2. Have the AI answer them using the resume as context
npx resumejson_cli ats_answering -i ./resume.json -q ./questions.unanswered.json -o ./questions.answered.json

# 3. Merge the answers back into a richer resume JSON
npx resumejson_cli ats_answered  -i ./resume.json -q ./questions.answered.json   -o ./resume.answered.json

ats_optimize

Rewrites the resume guided by an ATS review (typically run on the post-ats_answered resume).

npx resumejson_cli ats_optimize -i ./resume.answered.json -r ./ats_review.json -o ./resume.optimized.json

Workflow: full ATS pipeline

The recommended end-to-end pipeline, mirroring the full_pipeline script in package.json:

# 1. ingest
npx resumejson_cli from_pdf      -i resume.pdf                                  -o resume.json

# 2. baseline score (optional, for before/after comparison)
npx resumejson_cli ats_score     -i resume.json                                 -o resume.ats_score.json

# 3. enrich with answered ATS questions
npx resumejson_cli ats_question  -i resume.json                                 -o questions.unanswered.json
npx resumejson_cli ats_answering -i resume.json -q questions.unanswered.json    -o questions.answered.json
npx resumejson_cli ats_answered  -i resume.json -q questions.answered.json      -o resume.answered.json

# 4. review and optimize
npx resumejson_cli ats_review    -i resume.answered.json                        -o resume.ats_review.json
npx resumejson_cli ats_optimize  -i resume.answered.json -r resume.ats_review.json -o resume.optimized.json

# 5. render the optimized result
npx resumejson_cli to_pdf        -i resume.optimized.json -o resume.optimized.pdf
npx resumejson_cli to_markdown   -i resume.optimized.json -o resume.optimized.md

# 6. (optional) confirm the score improved
npx resumejson_cli ats_score     -i resume.optimized.json -o resume.optimized.ats_score.json

When the user asks to "improve / optimize my resume for ATS", run this full pipeline unless they ask for a single step.

Output

• All commands write structured JSON (or PDF / Markdown for conversion outputs) to the path given in -o.
• AI-backed commands also print a human-readable pretty-printed summary to stdout.
• Use - for either -i or -o to read from stdin / write to stdout.
• The from_pdf PDF→images step is memoised in .openai_cache.sqlite next to the package, so re-running it on the same PDF is cheap.

Error handling

• Missing OPENAI_API_KEY — every AI command will fail; tell the user to export it before retrying.
• Input JSON that does not match ResumeJsonSchema — the CLI throws a Zod validation error; surface it and ask the user to fix the offending field rather than silently retrying.
• from_pdf on a non-text / image-only PDF still works (it uses the vision model), but very long PDFs are slow and may need a smaller input.