thecolab-ai/.skills

Community-contributed AI skills for useful New Zealand-specific data and workflows.

Point your agent at real NZ infrastructure, public datasets, market data, industry feeds, transport APIs, pricing sources, weather services, and other useful local information.

What is this?

Think Folding@home, but for spare AI tokens. Instead of donating idle compute, the community contributes skills that make New Zealand-specific data actually usable by AI agents.

Every skill in this repo is a drop-in connector. Clone the repo, install the skill, and your agent gains access to real NZ-relevant data without you having to reverse-engineer the source.

Try it in 30 seconds

git clone https://github.com/thecolab-ai/.skills
cd .skills

# NZ fuel prices, supply, and vessels
python3 skills/fuelclock-nz/scripts/cli.py summary

# Latest NZ news
python3 skills/nz-news/scripts/cli.py headlines

# NZ supermarket stores and pricing history
python3 skills/grocer-nz/scripts/cli.py stores --query Papakura

# Auckland Transport real-time (needs AT API key)
python3 skills/at-transport/scripts/cli.py alerts

No auth is needed for fuelclock-nz or nz-news. at-transport requires a free API key from dev-portal.at.govt.nz, and metservice-nz requires a MetOcean API key.

Optional browser-assisted mode

Most skills use direct public HTTP/API calls. Some modern sites only expose useful public data reliably inside a real browser context, especially on headless servers. Those skills may offer an explicit --browser flag.

Recommended runtime: CloakBrowser. Full convention: docs/browser-assisted-skills.md.

Recommended convention:

• --browser should use CloakBrowser when it is installed.
• CloakBrowser must be optional: normal validation and non-browser smoke tests should still run on clean hosts.
• If CloakBrowser is missing, the CLI should return a clear cloakbrowser_not_installed error so the agent can recommend installing it instead of silently pretending browser mode ran.
• Browser-assisted skills must stay read-only and stop before login, checkout, payment, booking, cart mutation, or protected account actions.
• CAPTCHA, request-auth, and bot challenges are blocked states, not puzzles to defeat; return an explicit blocked flag or clear error and use a public fallback when one exists.

Access policy

Skills prefer official APIs, feeds, and downloads that respond to public, unauthenticated requests. The shared lib/nzfetch.py helper makes one direct request first and, when a user has configured a proxy, may make a bounded set of retries for blocked responses, rate limits, truncated reads, or recognised challenge pages. Proxy settings are optional secrets loaded from the environment; skills must never commit, print, or include them in errors.

If the bounded attempts cannot retrieve the public resource, the skill must return an explicit blocked or rate-limited outcome rather than an empty success. The repository does not provide CAPTCHA solving, authentication bypass, protected-token forgery, account operation, checkout, booking, payment, or transaction automation. Full contributor guidance is in CONTRIBUTING.md, and source operators can request removal, attribution, rate changes, or a different supported integration through COMPLAINTS.md.

About this repository

This repo is intentionally structured like a proper skills library.

It includes:

• skills/ for real contributed skills
• spec/ for the Agent Skills spec link plus our repo stance
• templates/ for the five source/workflow scaffold variants
• scripts/ for scaffolding, validation, contracts, pack installation, and catalogue generation
• skills.json and packs/ for generated discovery and trust-based installation manifests

We are borrowing the shape of the best public skills repos, but not copying their skill content.

Install a trust pack

The default nz-public-data pack contains only public read-only connectors. It excludes commercial websites, personal data, paid providers, internal publishing workflows, and artifact writers. Preview an installation before copying it into an explicit skills directory:

python3 scripts/install_pack.py nz-public-data \
  --target /absolute/path/to/agent-skills \
  --dry-run

python3 scripts/install_pack.py nz-public-data \
  --target /absolute/path/to/agent-skills

The six generated manifests and their trust models are documented in docs/packs.md. Existing users can continue using the legacy nz-skills aggregate while migrating to a narrower pack.

For businesses whose APIs we touch

If you operate a service that one of these skills connects to and you'd prefer we didn't, email [email protected] with the skill name and a brief reason. We action good-faith requests from the affected operator promptly, usually within 5 working days. Full process in COMPLAINTS.md.

But before you do, hear us out on why we built this.

These skills access endpoints that respond to unauthenticated public requests. Requests may use an optional user-configured proxy with bounded retries, as described in the access policy above. We don't bypass authentication, solve CAPTCHAs, forge protected tokens, or operate accounts and transactions. Users of these skills are responsible for complying with the terms of service of the underlying provider. What we've done is make public data legible to AI agents, so a Kiwi using an LLM can actually get useful answers about NZ fuel prices, bus times, pharmacy stock, or property data without us all reinventing the same wrapper a thousand times.

The thing is, agents are going to access this data either way. The only question is whether they do it well, with one clean shared connector, or badly, with ten thousand half-broken scrapers hammering your servers. We think one clean connector is better for everyone, you included.

If you're a business looking at this and thinking "how do we actually want AI agents to interact with our data" rather than "how do we stop them", that's a conversation worth having. We do that work. Email [email protected] or open an issue and we'll help you design the right surface: a proper agent-ready API, sensible rate limits, an auth model that fits, the lot. The businesses that figure this out early will own their corner of the agent economy. The ones that don't will get scraped anyway, just less efficiently.

Available skills

Skill	Trust pack	Source owner	Description
airnz-flights	nz-commercial-web	Air New Zealand	Search public Air New Zealand fare snapshots and timetable fallback data through a lightweight no-login CLI. Use when the task involves Air NZ route/date fare snapshots, fare products, flight numbers, departure/arrival times, duration, stops, or machine-readable current flight data. Optional --browser mode uses CloakBrowser when installed and returns cloakbrowser_not_installed when unavailable. Read-only; no login, Airpoints, seat holds, booking, payment, manage-booking, or checkout actions.
akahu-personal	nz-personal-data	Akahu	Query personal transactional bank-account data — account lists, balances, settled/pending transactions, account-specific transaction history, and private JSON/CSV exports — from Akahu, a New Zealand open-banking provider. Use when the task involves personal bank data exports, NZ account balances, transaction analysis, cashflow/spending categorisation, or inspecting authenticated Personal App banking data. Requires user-provided AKAHU_APP_TOKEN and AKAHU_USER_TOKEN; never smoke-test against real personal data in CI.
akl-housing-capacity	nz-public-data	Auckland Council	Query Auckland housing supply trends and 2022/23 plan-enabled and feasible housing capacity. Use for monthly housing updates, building consents, HBA, capacity for growth, PC78, local-board growth capacity, social housing delivery and housing demolition source discovery.
akl-ontrack	nz-public-data	AKLOnTrack	Read AKLOnTrack's public Auckland train observation archive. Use for daily stop-by-stop lateness, cancellations and skipped stops from data.aklontrack.com, not the live Auckland Transport API or a GTFS timetable.
akl-rainfall

Machine-readable filters, health, authentication, access mode, maintenance ownership, and outbound domains are in skills.json.

Skill sets

• ./skills - Real skills contributed to this repo
• ./spec - Agent Skills spec and repo-specific stance
• ./templates - Source/workflow scaffold variants
• ./packs - Generated trust-based installation manifests
• ./skills.json - Generated machine-readable catalogue

Skill Development Guidelines

All skills in this repository use Python as the canonical entrypoint.

• Entrypoint: Every skill must have scripts/cli.py with #!/usr/bin/env python3 as its first line.
• Python stdlib preferred. Use urllib.request, json, re, xml.etree.ElementTree, html.parser, argparse, and the standard library where possible. If a third-party dependency is truly necessary, declare it in a requirements.txt in the skill directory or via inline # /// script PEP 723 metadata.
• TypeScript/Node helpers are not accepted. jetstar-flights has a declared legacy exception behind a canonical Python wrapper; do not copy it.
• Every skill has: SKILL.md, scripts/cli.py, scripts/test_contract.py, scripts/smoke_test.py, and deterministic fixtures.
• Every data command exposes --json. Use scripts/run_skill.py when the versioned common result envelope is required.
• Stable failures: use the exit codes in docs/contracts.md and return blocked/schema states explicitly.

Contributing

We’ve made this repo opinionated on purpose.

Don’t freestyle the structure. Generate a scaffold, fill it in properly, then validate it before you open a PR.

Quick start

git clone https://github.com/thecolab-ai/.skills
cd .skills

python3 scripts/new_skill.py my-nz-skill \
  --variant public-api \
  --description "Query the official example API. Use for example records and status checks." \
  --source-owner "Example Agency" \
  --source-url "https://example.govt.nz/api"
python3 scripts/validate_skill.py skills/my-nz-skill
python3 skills/my-nz-skill/scripts/test_contract.py
python3 scripts/run_smoke_tests.py my-nz-skill
python3 scripts/generate_catalogue.py

Available scaffold variants

• public-api
• public-download
• html-readonly
• authenticated-personal
• documentation-workflow

Canonical skill folder structure

skills/
  my-nz-skill/
    SKILL.md
    references/source-notes.md
    scripts/
      cli.py
      test_contract.py
      smoke_test.py
    tests/fixtures/

See docs/contracts.md, docs/packs.md, and docs/migration.md.

Contribution rules

• Skills should target useful NZ-specific data or workflows
• Public and open data is great, but not the only valid source
• If auth is required, document it clearly in SKILL.md, including:
  • How to obtain access
  • Required environment variables
  • Whether there’s a free tier
  • Example .env entry
• Skills that scrape public websites or consume open RSS feeds should note rate limits, caching expectations, and source stability concerns
• description must say what the skill does and when to use it
• Keep SKILL.md lean and operational
• Move deep detail into references/
• Put deterministic helpers in scripts/cli.py using Python stdlib
• Do not add per-skill README.md, CHANGELOG.md, or junk notes

Full guide: CONTRIBUTING.md

Good areas to cover

• Government and public datasets
• Transport and logistics feeds
• Weather, marine, and environmental data
• Retail and pricing data
• Property, mapping, and local infrastructure data
• Industry-specific NZ information sources
• Other useful NZ-centric APIs or data workflows

Community

Built by the TheColab community, New Zealand's AI builders' collective.

Join us at the next weekly catchup.