AutoCCV

Turn a PDF academic CV into an import-ready Canadian Common CV (CCV) XML — driven by Claude Code.

Updating a CCV by hand through the web portal is painful for anything bulk: publications, presentations, supervision, funding. AutoCCV reads your PDF CV, extracts every entry, maps it to the correct CCV sections and controlled-vocabulary codes, enriches publications with DOIs, and emits a single XML file you import in one shot.

The intelligence (reading the PDF, deciding what each line is, asking you when unsure) runs inside Claude Code via a bundled skill. The XML generation, code resolution, cleaning, and validation run as deterministic, tested Python — so the output is reproducible and the importer-breaking gotchas (embedded line breaks, non-ASCII diacritics) are handled automatically.

Never heard of Claude Code? It's Anthropic's AI coding assistant that runs in your terminal and can drive this whole tool for you. You don't need to be a programmer. See New to Claude Code? (setup guide) below — or just skim the Quick start and follow along.

PDF CV ──▶  Claude Code (SKILL)  ──▶  cv_data.json  ──▶  autoccv (Python)  ──▶  CCV-output.xml
                 │                         ▲                    │
          asks you when unsure       the contract        seed CCV export (your identity)
          enriches DOIs (Crossref)                       + shipped catalog of CCV codes

• You provide: your PDF CV and a seed CCV export (your CCV with Personal Information already filled, exported as XML from the portal).
• Claude Code produces: cv_data.json (structured CV), then runs the Python pipeline to produce CCV-output.xml plus NOTES.md (assumptions and fields to double-check before importing).

Every CCV field/section/code ID is a global constant (verified across multiple real CVs), so the shipped catalog + your seed are all that's needed — no manual code lookups.

1. Fill Personal Information & export a seed. Log in at https://ccv-cvc.ca → CV → Generic → fill the Personal Information section → export the CV as XML. Save it into this repo (e.g. my-seed.xml).

2. Clone and open in Claude Code.

   git clone https://github.com/dvida/AutoCCV.git
   cd AutoCCV
   pip install -r requirements.txt
   claude            # open Claude Code in this directory

3. Point it at your CV. In Claude Code:

   Use the autoccv skill to convert My-CV.pdf into a CCV. My seed export is my-seed.xml.

   Claude Code reads the PDF, asks you a few grouped questions (publication status, co-author roles, ambiguous DOIs…), and runs the pipeline.

4. Review & import. Read NOTES.md, fix any flagged fields if you wish, then import CCV-output.xml at the portal (Utilities → Import). Keep your seed export as a backup.

AutoCCV requires the seed export — it guarantees your identity records are correct and lets it reuse organization IDs you've already entered.

Experienced Claude Code users can skip this — the Quick start above is all you need.

What is Claude Code? Claude Code is Anthropic's AI coding assistant that runs in your terminal (or VS Code / JetBrains). You talk to it in plain English; it reads files, runs commands, and edits code for you. For AutoCCV, it reads your PDF CV, asks you a few questions, and runs the conversion — you don't have to write any code.

Works on Windows, macOS, and Linux. The official setup guide has per-OS instructions and is the authoritative source if anything below differs — start there if you get stuck.

• Node.js 18+ — required by Claude Code. Download the LTS installer for your OS from https://nodejs.org (Windows .msi, macOS .pkg, or your package manager).
• Python 3.9+ — required by AutoCCV's pipeline.
  • Windows: install from https://www.python.org/downloads/ and tick "Add python.exe to PATH" in the installer. (Or winget install Python.Python.3.12.)
  • macOS: brew install python (or python.org). Linux: usually preinstalled, else sudo apt install python3 python3-pip.
• Git — to download this repo: https://git-scm.com/downloads (Windows: this also gives you "Git Bash", a handy terminal).

Terminal to use: Windows → Windows Terminal or PowerShell (or Git Bash); macOS → Terminal; Linux → your shell. On Windows, type python (not python3) — substitute that in any python3 … command below. Claude Code itself handles this for you when it runs the pipeline.

In your terminal (same on every OS):

npm install -g @anthropic-ai/claude-code

(See the official setup guide for native installers, IDE extensions, and Windows-specific notes if you hit snags.)

Run claude once and it will walk you through signing in. Two ways to pay for usage:

• A Claude subscription (simplest, flat monthly fee) — sign in with your claude.ai account:

  • Pro (~US$20/mo): fine for a one-off CV conversion.
  • Max (~US$100–200/mo): more headroom; nice if you'll iterate a lot or run other Claude Code work. A long CV uses a fair amount, but a single conversion is comfortably within Pro limits.

• API pay-as-you-go via the Anthropic Console — billed per token; good if you only run this occasionally and don't want a subscription.

  See current options on the pricing page and the Claude Code overview.

git clone https://github.com/dvida/AutoCCV.git
cd AutoCCV
pip install -r requirements.txt        # one-time: installs the Python dependency (lxml)
claude                                  # starts Claude Code in this folder

Then type a request like:

Use the autoccv skill to convert My-CV.pdf into a CCV. My seed export is my-seed.xml.

Put your My-CV.pdf and your my-seed.xml (see Quick start step 1) in the AutoCCV folder first so Claude Code can find them. It takes over from there — answer its questions, and you'll get CCV-output.xml + NOTES.md.

Handy in-session commands: type /help for help, /skills to see available skills (you should see autoccv), and /exit to quit. New to the interface? The quickstart is a 5-minute read.

• Claude Code overview & docs: https://docs.claude.com/en/docs/claude-code/overview
• Quickstart: https://docs.claude.com/en/docs/claude-code/quickstart
• Setup / install: https://docs.claude.com/en/docs/claude-code/setup
• Pricing & plans: https://www.claude.com/pricing

If you write cv_data.json yourself (see schema/cv_data.schema.json and examples/sample_cv_data.json), you can run the pipeline directly (on Windows use python instead of python3):

# 1. (optional) enrich publications with DOIs/volume/issue from Crossref
python3 -m autoccv.doi cv_data.json

# 2. generate the CCV XML from your seed + cv_data
python3 -m autoccv.generate --seed my-seed.xml --data cv_data.json --out CCV-output.raw.xml

# 3. make it import-safe (ASCII, single line)
python3 -m autoccv.clean CCV-output.raw.xml -o CCV-output.xml

# 4. validate
python3 -m autoccv.validate CCV-output.xml      # -> "OK — valid and import-safe"

(Installing the package via pip install . also gives the autoccv-build/-clean/-validate/-doi commands.)

.claude/skills/autoccv/SKILL.md   the Claude Code skill (the procedure Claude follows)
autoccv/                          deterministic Python package
  ccvgen.py      lxml primitives: clone records, set value/lov/refTable/date, insert at right depth
  generate.py    data-driven generator: cv_data.json + skeleton + section_map -> XML
  resolver.py    maps human labels -> CCV lov ids / refTable chains (exact -> ascii -> fuzzy)
  clean.py       transliterate non-ASCII + strip line breaks + single-line (the import-hang fix)
  validate.py    well-formedness + recordId + import-safety lint
  merge.py       harvest org IDs from the seed; dedup vs existing records
  doi.py         Crossref enrichment
data/                             shipped, generated from the example CVs
  skeleton.xml          one blank template record per section (carries real field IDs)
  section_map.json      cv_data field -> CCV field label + type + lov/refTable rule, per section
  lov_catalog.json      CCV field label -> {human label -> code id}
  reftable_catalog.json organizations / geography / research-classification trees
  section_paths.json    where each section lives in the container hierarchy
schema/cv_data.schema.json        the LLM -> generator contract
build_tools/                      maintainer scripts that (re)build data/ from examples/
examples/                         real CCV exports + a sample cv_data.json + a minimal seed
tests/                            pytest suite

The catalogs and skeleton are harvested from real CCV exports in examples/. To rebuild them (e.g. after adding a richer example with more sections):

cd build_tools
python3 extract_catalogs.py        # -> data/lov_catalog.json, data/reftable_catalog.json
python3 extract_section_paths.py   # -> data/section_paths.json
python3 extract_skeleton.py        # -> data/skeleton.xml   (run after section_paths)

The more (and more complete) example CVs you add to examples/, the richer the catalog of organizations, disciplines, and controlled-vocabulary options becomes.

Privacy: real CCV exports contain personal data, so examples/CCV-*.xml are git-ignored and not published. The committed data/ artifacts are derived and PII-free (the skeleton is blanked; the catalogs hold only organization/vocabulary names). Maintainers keep raw exports locally to rebuild data/. End users never need them — data/ ships pre-built.

Degrees · Academic & Non-academic Work Experience · Research Funding · Courses Taught · Student/Postdoctoral Supervision · Committee & Other Memberships · Recognitions · Editorial & Journal Review Activities · Credentials · Presentations · Journal Articles · Conference Publications · Reports · Books · Book Chapters · Text & Broadcast Interviews · Knowledge & Technology Translation · International Collaboration Activities · Event Administration · Organizational Review Activities · Leaves of Absence and Impact on Research · Affiliations · Working Papers.

Personal Information comes from your seed. Adding a new section is a JSON edit in data/section_map.json (plus a skeleton record harvested from any export that has it) — no code change.

• Controlled vocabulary — degree types, funding/supervision roles, recognition types, organizations, research disciplines, etc. resolved to the exact CCV code IDs; unresolved values fall back to free text and are flagged in NOTES.md.
• Import-hang prevention — the CCV importer stalls on embedded line breaks and non-ASCII characters; clean removes both and validate re-checks.
• DOI enrichment — missing DOIs/volume/issue filled from Crossref with confidence-gated auto-accept.
• Idempotent re-runs — records already in your seed are de-duplicated, so you can iterate.

pip install -r requirements.txt
python3 -m pytest -q

AutoCCV is an unofficial tool and is not affiliated with the Canadian Common CV / CIHR. CCV field and code identifiers were observed from real exports and may change if the CCV schema is updated; re-run the build_tools extractors against a fresh export if imports start failing. Always review NOTES.md and keep a backup of your existing CV before importing.

MIT — see LICENSE.