NZ regulatory analysis

Goal

Discover current official regulatory-analysis publication pages and inspect their linked PDFs. The skill reads the Ministry for Regulation's RAS index and the Ministry for the Environment's Cabinet-paper/RIS index. It classifies records without downloading or interpreting PDF contents.

Use this when

• Finding current official RAS, RIS, RIA, supplementary analysis reports, or cost recovery impact statements.
• Locating independent, departmental, agency, or otherwise-labelled quality assessments.
• Finding the Cabinet papers and minutes linked from an environmental regulatory package.
• Producing source URLs and retrieval timestamps before a policy comparison.

Do not use this for

• Recovering deleted, superseded, or archived publications.
• Claiming that a search result is the complete government-wide corpus.
• Treating a document's presence as proof that its analysis or quality assurance is adequate.
• Summarising PDF contents: this CLI returns links and labels but does not fetch PDFs.
• Circumventing CAPTCHA, access challenges, robots controls, or request blocks.

Preferred workflow

1. Run search with a narrow policy phrase. Use --source when the publisher is known.
2. Select the official publication-page URL from the results.
3. Run inspect URL --json to list linked official PDFs and their distinct document types.
4. Open and read the relevant returned PDF separately before making content claims.
5. Cite the exact page/PDF URL and source.retrieved_at; explain that listings can change.

CLI

Run from the repository root:

python3 skills/nz-regulatory-analysis/scripts/cli.py <command> [flags]

Commands:

• search QUERY [--source all|regulation|environment] [--limit 1-25] [--json] searches one or both current indexes. QUERY is positional.
• inspect URL [--json] fetches one supported official HTML publication page and lists same-site linked PDFs. URL is positional.

Examples:

python3 skills/nz-regulatory-analysis/scripts/cli.py search "climate disclosures" --json
python3 skills/nz-regulatory-analysis/scripts/cli.py search freshwater --source environment --limit 5
python3 skills/nz-regulatory-analysis/scripts/cli.py inspect "https://www.regulation.govt.nz/publications-and-resources/regulatory-analysis-summaries/EXAMPLE/" --json

Result contract

--json emits the repository result envelope (schema_version: "1"). Successful live results include source.url, source.retrieved_at, query, data, warnings, and blocked: false.

• Search records retain title, official url, displayed publication date, publisher, and document_type; source-specific author/tags are retained when shown. Regulation searches follow all result pages within the safety bound so source_totals.regulation is not a page-one count. Multi-document environment packages use publication_package and expose every labelled type in document_types rather than collapsing Cabinet and regulatory documents into one type.
• Inspect results retain the page title/type/update value plus deduplicated documents with official PDF URLs and separate document types.
• A genuine no-match search is ok: true with results: [].
• HTTP/access challenges are exit 4 with blocked: true; upstream outages are exit 5; changed/malformed page structures are exit 6. They never become empty success.

Document types include regulatory_analysis_summary, regulatory_impact_statement, regulatory_impact_assessment, supplementary_analysis_report, cost_recovery_impact_statement, quality-assessment variants, Cabinet paper/minute, briefing, and conservative fallback types.

Resources

• scripts/cli.py — bounded stdlib HTTP client and parser.
• scripts/test_contract.py — deterministic repository contract runner.
• tests/test_parser.py and tests/fixtures/ — scrubbed minimal parser and adversarial cases.
• scripts/smoke_test.py — fixture checks plus bounded normal/empty/blocked-aware live contracts.
• references/source-notes.md — source surfaces, limits, and parser assumptions.

Safety and limits

• Read-only GET requests, no authentication, browser automation, cache, proxy configuration, or PDF download.
• Regulation searches follow at most 10 index pages (up to 120 results with the current 12-result page size); a broader result set fails explicitly and asks for a narrower query rather than reporting a partial total. Environment search makes one index request. Inspect makes one request. Each request has a 10-second timeout and 2 MB HTML cap.
• Redirects remain inside the two declared official hosts and supported publication paths.
• Labels drive type classification; inspect the document itself before relying on that classification in consequential work.