Innblikk logo

Innblikk

Organization
navikt
innblikk

Sett opp Innblikk (Team ResearchOps' nettstatistikk, basert på Umami) i en Nav-app som ikke bruker dekoratøren. Bruk når et team vil telle sidevisninger med sporing.js, trenger samtykke etter ekomloven, CSP for Innblikk, en before-send-hook som holder private sider ute, eller vil sjekke en eksisterende Innblikk-integrasjon. Bruker appen dekoratøren, bruk nav-dekoratoren i stedet.

Overview

Publishernavikt
Repositorycopilot
Skill nameinnblikk
Stars
60
Forks
16
Bundled files
1
LicenseMIT
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 1 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by navikt on GitHub. Read the source before you install it.

Installation

Install the Innblikk AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/navikt/copilot/tree/main/skills/innblikk
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/navikt/copilot.git /tmp/copilot
mkdir -p .claude/skills
cp -r /tmp/copilot/skills/innblikk .claude/skills/innblikk
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Innblikk in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Innblikk on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Innblikk is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Innblikk without dekoratøren

Innblikk is Team ResearchOps' web analytics, based on Umami. It was called Umami Start before. Apps that use Nav Dekoratøren already get Innblikk through it, with consent handled. For those, use the nav-dekoratoren skill instead. This skill is for apps that load the script themselves.

Slack: #researchops Worked example: apps/my-copilot in navikt/copilot (ki-utvikling.nav.no)

Step 1: Tracking code and approval

  1. Get a tracking code (website ID) from https://innblikk.ansatt.nav.no/sporingskoder, or ask in #researchops on Slack.
  2. Before prod: fill in the reusable Innblikk etterlevelsesdokument (linked from the "Kom i gang" page on innblikk.ansatt.nav.no) and describe the use on the app's privacy page.

The website ID is public: it ends up in the page HTML. Pass it in as a runtime environment variable, so it can be switched on and off without a rebuild:

yaml
# .nais/app.yaml (Handlebars template, values per environment)
env:
  {{#if innblikk_website_id}}
  - name: INNBLIKK_WEBSITE_ID
    value: "{{innblikk_website_id}}"
  {{/if}}

Render nothing when the variable is unset.

Step 2: The script tag

Use this exact URL: https://cdn.nav.no/team-researchops/sporing/sporing.js.

html
<script
  defer
  src="https://cdn.nav.no/team-researchops/sporing/sporing.js"
  data-website-id="YOUR-WEBSITE-ID"
  data-domains="minapp.nav.no"
  data-exclude-search="true"
  data-do-not-track="true"
  data-before-send="innblikkBeforeSend"
></script>
AttributeEffect in sporing.js
data-website-idThe tracking code. No ID, no sends.
data-domainsComma-separated hostnames. On any other host nothing is sent. Set only the prod host, so dev and preview send nothing.
data-exclude-search="true"Strips the query string from the URL and referrer. The hash is kept.
data-do-not-track="true"Respects the browser's Do Not Track setting.
data-before-sendName of a function on window. It gets (type, payload) before every send. Return the payload to send it, or a falsy value to cancel.

The script also reads data-tag, data-auto-track, data-fetch-credentials (default omit) and data-opt-out-filters. Do not use attributes beyond these. It ignores them.

Behaviour to know before writing the hook:

  • Page views are sent on load, and about 300 ms after a history.pushState/replaceState that changes the URL, so client-side navigation is tracked once the script is loaded. Back and forward (popstate) are not tracked.
  • On navigation, referrer is the previous URL in the app. On first load it is document.referrer, or empty when that is the same origin.
  • The hash stays in url and referrer.
  • Clicks on elements with data-umami-event or data-sporing-event attributes send custom events; every data-umami-event-<key> attribute on the element is sent as event data, and the hook below passes name and data through unchanged. Never put personal data in them, and do not add the attributes at all if the app counts page views only.

The before-send hook

Define it on window before the script loads. sporing.js looks up window[name] at every send, which is why withdrawn consent takes effect without a reload. It must drop private pages, drop private referrers and strip the hash:

ts
export function innblikkBeforeSend(_type: string, payload: { url?: string; referrer?: string }) {
  if (readConsent() !== true) return false; // withdrawn consent stops sends without a reload
  const path = (u: string) => decodePath(new URL(u, location.href).pathname);
  if (!payload.url || isPrivatePath(path(payload.url))) return false;
  const noHash = (u?: string) => u?.split("#")[0];
  const referrer = payload.referrer && !isPrivatePath(path(payload.referrer)) ? noHash(payload.referrer) : undefined;
  return { ...payload, url: noHash(payload.url), referrer };
}
if (typeof window !== "undefined") {
  (window as unknown as Record<string, unknown>).innblikkBeforeSend = innblikkBeforeSend;
}

decodePath is decodeURIComponent with a fallback, so /%61dmin counts as /admin. Use the same private-path list that your access control enforces, so a new protected page is excluded automatically.

Step 3: Consent

Analytics storage and reading on the user's device needs consent under ekomloven § 3-15. This is the reason for the flow below, not legal advice. Ask your privacy contact if in doubt.

  • Load nothing before an explicit yes: no script tag, no request to cdn.nav.no.
  • Show a yes button and a no button with the same size and style.
  • Store the choice locally with a version and a date, for example in localStorage as { analytics: true, version: 1, at: "2026-10-31T09:00:00.000Z" }. Treat an unknown version as no choice.
  • On no or withdrawal, set both umami.disabled and sporing.disabled in localStorage. sporing.js checks both before every send. On yes, remove them.
  • sporing.js also checks the flags once at load. If a flag is set then, it never installs its navigation hooks, so remove the flags before loading the script, as setConsent does.
  • Make withdrawal as easy as consent, for example with the same two buttons on the privacy page.
ts
export function setConsent(analytics: boolean) {
  localStorage.setItem(CONSENT_KEY, JSON.stringify({ analytics, version: 1, at: new Date().toISOString() }));
  for (const flag of ["umami.disabled", "sporing.disabled"]) {
    if (analytics) localStorage.removeItem(flag);
    else localStorage.setItem(flag, "1");
  }
  window.dispatchEvent(new Event(CONSENT_EVENT));
}

The reference (innblikk-consent.ts) wraps the storage calls in try/catch. Do the same: if storage throws, keep the choice in memory for the page load so the banner does not come back on every render.

Step 4: CSP and Referrer-Policy

text
script-src  'self' https://cdn.nav.no/team-researchops/sporing/sporing.js
connect-src 'self' https://reops-event-proxy.nav.no
  • Allow the exact script URL, not all of cdn.nav.no.
  • sporing.js posts to https://reops-event-proxy.nav.no/api/send. On hosts containing .dev.nav.no or .dev.intern.nav.no it posts to https://reops-event-proxy.ekstern.dev.nav.no instead. Add that host to connect-src in dev only if dev should send.
  • Set Referrer-Policy: strict-origin-when-cross-origin, so the full URL with query string is not sent in the Referer header to cdn.nav.no and the event proxy.

Step 5: Data rules

  • Page views only.
  • No personal data in events, page titles or query strings. The title is sent with every page view.
  • Never load the script on logged-in or admin pages, and cancel events from them in the hook.
  • Do not call identify() (window.umami.identify / window.sporing.identify). The id is attached to every later event.
  • Tracking free-text or search content is out of scope here. Ask #researchops and your privacy contact first.

Framework notes

Next.js App Router

  • A client component ("use client") renders the banner or the script with next/script (strategy="afterInteractive").
  • Read the choice with useSyncExternalStore and an undefined server snapshot, so the server HTML has neither the banner nor the script.
  • Read process.env.INNBLIKK_WEBSITE_ID in a server component that renders per request, and pass it as a prop. A NEXT_PUBLIC_ variable would be baked in at build time.
  • Assign window.innblikkBeforeSend at module scope in the client component, so it exists before sporing.js runs. Guard it with typeof window !== "undefined", because the module also runs on the server.
  • Set CSP and Referrer-Policy in headers() in next.config.ts.

Plain HTML and other frameworks

Define window.innblikkBeforeSend first. After a yes, inject the tag:

js
const s = document.createElement("script");
s.src = "https://cdn.nav.no/team-researchops/sporing/sporing.js";
Object.entries({
  websiteId: "YOUR-WEBSITE-ID",
  domains: "minapp.nav.no",
  excludeSearch: "true",
  doNotTrack: "true",
  beforeSend: "innblikkBeforeSend",
}).forEach(([k, v]) => (s.dataset[k] = v));
document.head.appendChild(s);

Step 6: Verify

Headless browser (Playwright or similar) against prod or a host listed in data-domains. Record every request:

CaseExpected
Before a choiceNo request to cdn.nav.no or reops-event-proxy
After yessporing.js returns 200, POST /api/send returns 2xx (201 in a headless check on ki-utvikling.nav.no, 2026-10-31)
After no, then navigateNo new /api/send
Private page, consent givenNo script tag, no /api/send

Unit tests with controls. Every negative test needs a positive control in the same setup, or it passes when nothing renders at all. Examples from my-copilot:

  • No ID renders nothing; control: the same page with an ID renders the script.
  • No sends from a private page; control: a public page passes the hook.
  • No sets the opt-out flags and never loads the script; control: yes loads it.
  • The hook strips the hash; control: a plain URL is unchanged.
  • A private referrer is dropped; control: a public referrer is kept.

Mock next/script as a plain <script> so the render is observable.

Worked example: my-copilot

In navikt/copilot:

  • apps/my-copilot/src/components/innblikk.tsx: script tag, banner, innblikkBeforeSend, the withdrawal control for the privacy page.
  • apps/my-copilot/src/lib/innblikk-consent.ts: consent store and opt-out flags.
  • apps/my-copilot/src/lib/private-paths.ts: the private paths shared with access control.
  • apps/my-copilot/src/components/site-shell.tsx: passes INNBLIKK_WEBSITE_ID at runtime.
  • apps/my-copilot/next.config.ts: CSP and Referrer-Policy.
  • apps/my-copilot/.nais/app.yaml, prod-gcp.yaml, dev-gcp.yaml: how the ID gets in.
  • apps/my-copilot/src/components/innblikk.test.tsx: tests with controls.

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Innblikk AI skill do?

Sett opp Innblikk (Team ResearchOps' nettstatistikk, basert på Umami) i en Nav-app som ikke bruker dekoratøren. Bruk når et team vil telle sidevisninger med sporing.js, trenger samtykke etter ekomloven, CSP for Innblikk, en before-send-hook som holder private sider ute, eller vil sjekke en eksisterende Innblikk-integrasjon. Bruker appen dekoratøren, bruk nav-dekoratoren i stedet.

Why use Innblikk on TypingMind?

Because you install it once and use it with any model. Innblikk is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Innblikk in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/navikt/copilot/tree/main/skills/innblikk. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Innblikk?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Innblikk?

As many as you like. As long as a model supports skills, you can use Innblikk with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Innblikk AI skill free?

Yes. It is published on GitHub by navikt under the MIT license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇