Skip to content
Sumisura

Quickstart

Sumisura runs on your machine. Nothing is hosted, there is no account, and it calls the Claude API with your own key.

What you need

Section titled “What you need”
RequirementWhy
Docker + Docker ComposeRuns the web app — one container from a published image
An Anthropic API keySelection, Rewrite, cover letters and research calls (Claude API, pay per use)
Claude CodeRuns the tailor-cv skill, which drives the tailoring pipeline
Typst on your PATHRenders the PDF
Python 3.10+ (python3, standard library only)The tailor-cv skill’s prose clean-up step (clean-user-facing-text skill, downloaded on first use)
pdftotext (poppler-utils)Optional — the ATS-parsability check

1. Get it

Section titled “1. Get it”

From a published image (recommended). No checkout, no Node, no Go — one image with the backend and the frontend together, on one port (ADR-0039):

Terminal window
mkdir sumisura && cd sumisura
curl -O https://raw.githubusercontent.com/gio-del/sumisura/main/docker-compose.release.yml
curl -o .env https://raw.githubusercontent.com/gio-del/sumisura/main/.env.example

That .env already names this release — uncomment SUMISURA_VERSION=v0.4.0 in it to pin that image rather than following latest, so upgrades stay deliberate. You still need a checkout for the tailor-cv skill, which runs in Claude Code — but not to run the app.

From source, if you want to change the code, or want the dev server’s hot reload:

Terminal window
git clone https://github.com/gio-del/sumisura.git
cd sumisura
git checkout v0.4.0   # or track main

2. Add your API key

Section titled “2. Add your API key”
Terminal window
cp .env.example .env
# then set ANTHROPIC_API_KEY=sk-ant-...


# from a checkout:
cp -r data/examples/. data/


# from a published image, fetch the stubs instead:
# curl -L https://github.com/gio-del/sumisura/archive/refs/heads/main.tar.gz \
#   | tar -xz --strip-components=2 '*/data/examples'

That copy seeds data/ with stub Master Data to edit. Everything it writes — profile.yaml, experience/, projects/, cover-letter-snippets/ — is gitignored on purpose: your name, your employer and your clients never go near a commit (ADR-0038). The tracked copies under data/examples/ stay as examples.

Without a key the app still starts and you can browse and edit Master Data — only the calls to Claude fail.

3. Start the app

Section titled “3. Start the app”
Terminal window
docker compose up

From a published image, everything is on one port:

Terminal window
docker compose -f docker-compose.release.yml up
  • App and API: http://127.0.0.1:8080

From a checkout, docker compose up runs the development pair instead:

  • Backend: http://127.0.0.1:8080
  • Frontend: http://127.0.0.1:5173

Both are bound to localhost with no authentication, on purpose. See LAN-reachable mode if you want to reach it from your phone.

4. Replace the stub data with your own

Section titled “4. Replace the stub data with your own”

The copy in step 2 filled data/ with stub content for a fictional “Jane Doe” so the app has something to show on first run. Replace it with your own career history — every file below is untracked, so edit freely:

  • data/profile.yaml — contact details, education, publications, awards, languages
  • data/experience/*.md — one file per job or client engagement
  • data/projects/*.md — one file per project
  • data/cover-letter-snippets/*.md — optional reusable cover-letter paragraphs

You can edit these in the web app, or in your editor — they are plain Markdown with YAML frontmatter, and both paths write the same files. See Writing your Master Data for the shape of each file.

5. Install the tailoring skill

Section titled “5. Install the tailoring skill”

The pipeline is a Claude Code skill served from this repo’s own plugin marketplace:

Terminal window
claude plugin marketplace add .
claude plugin install sumisura@sumisura-local

6. Tailor your first CV

Section titled “6. Tailor your first CV”

In Claude Code, from the repo:

/sumisura:tailor-cv

Give it a job description (pasted text or a URL), point it at an Application you already track, or give it nothing at all for a general-purpose CV. It walks Selection → Rewrite → Text Review (you approve) → Render → Visual Review (you approve), and writes everything into output/<label>-<timestamp>/.

That directory holds cv.pdf, cover-letter.pdf, the assembled data.json, and selection.json — the record of which Entries were chosen and how each bullet was rewritten.

Next

Section titled “Next”