Methodology

PunkSolid methodology is how we keep agentic coding productive without pretending agents are approvers. The same patterns show up on client deliveries and on the PunkSolid ai-network sites in this repository: specs first, human gates on anything that affects trust or compliance, automated checks that block sloppy merges, and migration discipline when URLs and tracking cannot break.

Agents and human gates

We break work into specs, implementation tasks, and verification steps. Agents draft components, content, and refactors in Cursor; they do not silently ship to production.

Human review gates cover claims, tone, accessibility, brand fit, and regulated copy on patient-facing or financial pages. On migrations, editors and stakeholders sign URL inventories and redirect maps before cutover—no DNS switch until those lists are approved. Launch timing and rollback decisions stay with people who own traffic and compliance risk.

Optional 1:1 Cursor coaching helps your maintainers after handoff; code ownership is the default outcome—full repository access whether or not you book more work with us.

Guardrails and the quality gate in this repo

On client codebases we document organization rules and quality bars up front so page five thousand matches page one—the same discipline described on healthcare WordPress rebuilds before bulk generation.

Inside this repository, the ai-network package enforces a quality gate agents and humans must respect before pages can graduate from draft to reviewed:

• src/validate/quality.ts defines minimum body words per page kind, answer length on commercial pages, FAQ counts, and service steps/deliverables/engagement blocks.
• pnpm validate and pnpm ai-network:preflight run validation, tests, and registry parity so broken frontmatter or topic ownership cannot merge unnoticed.
• Metadata lint, orphan checks, and post-build link scans reduce shipping broken internal links on static deploys.

Those checks mirror what we recommend client repos adopt: types, lint, link scans, and human-readable specs—not “trust the agent diff.”

Content model replacements (ACF and WP All Import)

Approved migration facts: eight healthcare provider websites, each 1,000–6,000 pages, rebuilt from WordPress + Elementor, ACF, and WP All Import into code built in Cursor.

Teams relied on ACF for structured fields and WP All Import for bulk creates and updates. Each site received custom in-codebase replacements: structured content fields instead of ACF, and bulk page import paths instead of WP All Import. Editors maintain thousands of pages in Cursor with the same guardrails used during the rebuild.

That pattern generalizes on website migration and WordPress to Astro or Next.js engagements whenever a CSV or admin import used to be the editorial bottleneck.

Migration safety: URLs, redirects, and tracking parity

A faster site that loses calls, leads, or consent records is not an improvement. Methodology on migrations insists on:

• URL inventory reconciled with analytics and Search Console coverage.
• Redirect maps tested on staging for high-traffic paths.
• Metadata and structured-data parity compared pre- and post-launch.
• Tracking parity for the stack each client already uses. On the healthcare rebuilds that included CallTrackingMetrics, Google Tag Manager, GA4, Usercentrics, Microsoft Clarity, VWO, Jotform HIPAA forms, a custom chat widget, and a HIPAA-aware lead endpoint into each client’s CRM—carried across launches, not reinstalled from memory.

Post-launch we watch Search Console for crawl and indexing regressions during the first review window. Cutover docs include rollback notes and who can revert DNS or reverse-proxy rules.

Read the website migration SEO checklist for verification steps we embed in delivery.

Performance budget and how we report scores

Performance work has a budget: ship templates that respect real content weight—scripts for consent, testing, call tracking, and forms included—not empty lab fixtures.

On the eight healthcare sites, mobile PageSpeed scores were roughly 65–75 before and 92–98 after; desktop PageSpeed scores roughly 80–95 before and 92–98 after. Those numbers are PageSpeed scores from Lighthouse lab runs (0–100), not Core Web Vitals. Field Core Web Vitals were not passing before launch; in Search Console they passed on mobile and desktop about four days after launch—a separate signal from lab scores.

We keep that distinction in stakeholder reports so a high lab score is not mistaken for a CWV pass. Traffic and leads held steady or grew on those projects; confidential figures stay off this page.

Automated QA on deliveries

Beyond the ai-network gate, client repos should run linting, type checks, link scans, and smoke tests on critical templates and forms. Visual regression on hero and checkout templates is valuable when marketing runs frequent experiments.

Functional smoke tests cover checkout or HIPAA-aware forms when present. Agents may generate tests, but humans choose which paths are release blockers.

SEO parity after replatforming

Metadata, canonicals, hreflang when present, structured data, and sitemap coverage are compared pre- and post-launch. Internal links are scanned so orphan routes do not multiply after import.

When mirrors or multi-site networks exist—as in this monorepo—canonical ownership and absolute owner links stay explicit so preview hosts do not compete with production URLs.

How methodology connects to services

• Website migration applies this playbook end to end on large replatforms.
• Code ownership states what clients keep when methodology-driven delivery ends.
• Healthcare WordPress rebuilds is the reviewed proof story for guardrails, content-model replacements, tracking parity, and PageSpeed score ranges.

When you want production work scoped with these gates, start on services or the contact form with CMS, page count, and integrations that cannot break.