internetnl-cli

A command-line client for the Internet.nl batch API, plus a documented recipe for running your own batch instance.

Internet.nl tests a domain for IPv6, DNSSEC, RPKI, TLS and a set of informational checks. The public site tests one domain at a time in a browser; this tool submits a whole fleet to a batch API v2 endpoint, polls until the run finishes, and renders the result as a diffable table or as JSON for pipelines.

Status: implemented. The user-visible surface stays pinned in openspec/changes/archive/2026-09-08-add-internetnl-cli/design.md; the commands below match the current behaviour.

New here? See the illustrated how-it-works page (the whole flow, the commands, the deploy topology), or the runnable internetnl-cli-demo (quickstart + example hosts + CI-gate example).

Try it in your browser, no account needed: https://mwest2020.github.io/internetnl-cli-demo/ — type a domain, get a real result from the same v1.0.0 service every tenant uses, one run at a time (strictly rate-limited; see docs/reference/demo-api.md for the contract that page relies on and docs/how-to/demo-run.md for how it is run).

The rule

Measure your own hosts freely. Measure someone else's only where the measurement is one their operator already invites.

Internet.nl is a public service whose own website invites anyone to test any domain, and the Dutch government publishes a twice-yearly measurement of some 12,000 government domains, per domain, by name. A single run against a domain that already sits in that category stays inside the norm the service itself sets. A run against a domain that does not, without asking, does not.

The rest is on you, and the tool enforces none of it:

• Once, not repeatedly. A batch run costs the measured host real work — DNS lookups, TLS handshakes, connection attempts. One run answers your question. A loop is abuse, whoever owns the domain.
• A verdict is not an audit. Batch results differ from the website's in documented ways. Never present a score about someone else's domain as a finding about their security.
• Aggregate, don't shame. If you measure a whole sector, publish what most often goes wrong and how to fix it — not a ranking of who scored worst. The first is useful to everyone; the second only burns the relationship you presumably wanted.

Quickstart

export INTERNETNL_ENDPOINT=https://batch.internet.nl/api/batch/v2
export INTERNETNL_CREDENTIAL=you:…                     # hosted API needs an account; user:pass, split on the first ':'
internetnl submit --file hosts.txt                     # prints request-id, then polls
internetnl poll <request-id>                           # resume after a closed laptop
internetnl results <request-id> --json > results.json  # machine-readable

There is no default endpoint: you configure where you measure — the hosted instance (account required) or your own. Switching between them is a change of INTERNETNL_ENDPOINT, nothing else.

Getting a credential

Every endpoint needs one, and there are three routes:

1. A supporter key for api.westerweel.work — the netnl facade in front of an instance run by this project's author. A one-off donation via Buy Me a Coffee of $2 or more mints a lifetime credential and mails it to you automatically, usually within a minute. It is a beta service on a best-effort, no-SLA basis, with the same fair-use rate limit for everyone — read docs/how-to/supporter-key.md for exactly what that promises and what it does not.
2. An account on the hosted batch API at Internet.nl, if your organisation qualifies for one.
3. Your own instance — see Self-hosting below.

No credential at all is needed to try the tool: the browser page runs one domain at a time against the same facade, anonymously — the same service, not a scaled-down imitation of it.

What it is, and is not

• A client. The tests, the scoring and the verdicts belong to Internet.nl and run on the endpoint you point at. Nothing is measured, approximated or filled in locally: if the API is unreachable, the answer is an error, not a guess.
• A pipeline gate. --fail-on-scored exits non-zero when a subtest that counts toward the score fails, with an allowlist file for accepted exceptions. Informational findings are shown but never gate.
• Honest. Every result carries the endpoint host, the run timestamp and the API version. A subtest missing from a response renders as unknown, never as passing.
• Not a dashboard (upstream has Internet.nl-dashboard), not a scraper of the website UI, and not a reimplementation of any test.

Exporting for another tool

internetnl results <request-id> --format findings [--findings-out FILE] writes netnl-findings/v1: every subtest of a finished batch, verbatim — the API's own status and verdict, category names derived from the API's own results.categories, and the domain's own score and report URL. No aggregation, no re-scoring, no HTML scraping. It exists so a consumer (Wanderer is the first) can adopt internet.nl results as its own dimension without reimplementing internet.nl's category rules or the batch API's edge cases. --json and the table are unchanged; see docs/netnl-findings-v1.md for the schema.

Batch results are not website results

Upstream documents the differences: the connection test is unavailable, DNSSEC tests skip the registrar lookup, and no prechecks run on whether a hostname resolves at all. Never quote a batch verdict as "the internet.nl score" without naming the endpoint — which is why every result is labelled.

Configuration

Everything is environment-tunable; see openspec/changes/archive/2026-09-08-add-internetnl-cli/design.md for the full table (INTERNETNL_ENDPOINT, INTERNETNL_USERNAME, INTERNETNL_PASSWORD, timeouts, poll interval and maximum, batch size, config path). INTERNETNL_CREDENTIAL is a single username:password alternative to the username/password pair, split on the first :; set either form, never both. Credentials come from the environment only — never from command-line arguments, and they never appear in output or logs.

Use in CI

internetnl is meant to run in someone else's pipeline: --json, stable exit codes (0 ok, 1 config, 2 transport/API, 3 --fail-on-scored gate, 4 poll timeout) and a bundled GitHub Action (action.yml):

- uses: MWest2020/internetnl-cli@v1
  with:
    hosts: example.org
    endpoint: https://api.westerweel.work
    credential: ${{ secrets.INTERNETNL_CREDENTIAL }}

See docs/how-to/ci.md for the full input reference, a plain-CLI recipe for other CI systems (GitLab CI et al.), and the gate's exit-code semantics.

Self-hosting

If you cannot get an account on the hosted batch instance, and a supporter key does not fit, you can run your own. Read docs/reference/self-hosted.md first: it needs a server with a fixed public IPv4 address and IPv6 — not something you run behind NAT — and it is a maintained service, not a script. docs/how-to/deploy-instance-vps.md walks the whole deployment, and docs/how-to/self-hosting-pitfalls.md collects the five traps that cost this project days — including a documented setting that silently kills all container egress, and a certbot in the webserver image that cannot start, so certificates never renew.

Development

• Python via uv, standard library only; uv run pytest runs the suite (no test touches the network or your real $HOME).
• The repo is built through the habitat agent chain: role definitions live in .claude/agents/, and scripts/verify.sh is the non-negotiable gate a builder run must pass.

License

MIT. This licence covers the client only: the tests, the scoring and the Internet.nl name belong to the Internet.nl / Platform Internetstandaarden project and are theirs.