CLAUDE.md — peptide.diy

Guidance for Claude Code working in this repo. Read before making changes.

What this is

peptide.diy — a Jekyll static site (research-led peptide protocols, guides, glossary, and a reconstitution calculator), hosted on GitHub Pages (FletcherPoole/peptide.diy) at the apex custom domain peptide.diy. DNS is on Namecheap (registrar-servers.com nameservers). British/en-GB audience and spelling.

Deploy / build

Push to main → GitHub Pages auto-builds and deploys. There is no separate CI.
Check build status: gh api repos/FletcherPoole/peptide.diy/pages/builds/latest -q '"\(.status) | \(.error.message // "none")"'. A failed build does not take the site down — Pages keeps serving the last good build.
Local Jekyll build is usually NOT available on the dev machine (no Ruby/Jekyll). Validate what you can at source level (JSON/JS via node --check, hand-review Liquid) and rely on the Pages build to confirm.
Outage gotcha: if the site 404s with GitHub’s “Site not found”, the Pages custom-domain binding likely dropped to cname: null even though the repo CNAME file is intact. Fix: gh api -X PUT repos/FletcherPoole/peptide.diy/pages -f cname=peptide.diy, then re-enable https_enforced=true after the cert re-issues.

Architecture

Jekyll, config in _config.yml (plugins: jekyll-sitemap). Site URL https://www.peptide.diy.
_layouts/default.html — the only layout: gtag + Consent Mode, <head> includes, skip-link, nav, mobile-nav, <main>, footer, search modal, cookie-consent, deferred JS.
_includes/ — head.html, seo.html, structured-data.html (JSON-LD: Organization/WebSite/Breadcrumb/Article/WebApplication/FAQPage/HowTo), nav.html, mobile-nav.html, footer.html, search-modal.html, cookie-consent.html, calc-cta.html, protocol/guide funnels.
css/styles.css — single stylesheet, dark theme via CSS custom properties (:root). Cache-busted by ?v=2026-07-23 19:59:51 +0000.
js/ — app.js (nav dropdowns, mobile menu, FAQ accordions, the calculator, cookie consent), search.js, beehiiv-form.js (newsletter iframe), contact-form.js (Formspree).
_data/ — faq.yml, glossary_terms.yml.
Content pages are top-level .html with YAML front matter (protocols, guides, policies).
_drafts/ — working drafts; Jekyll ignores these in the build (never published). Contains _PROTOCOL-TEMPLATE.md (the canonical protocol spec) and in-progress *-draft.md outlines.

The calculator (js/app.js)

Reconstitution calculator on calculator.html (full) and index.html (home, injection-only). Features: injection/cream/nasal modes, live-updating results, aria-live region, URL-param deep-linking + persistence (state is written to the URL so results are shareable/bookmarkable), copy-link and save-as-PNG (canvas, no external libs), a barebones vial-duration estimate, and an SVG syringe fill visual. Protocol pages deep-link into it via the calc: front-matter block → calc-cta.html. GA events: calculator_used, calculator_share_link, calculator_save_image, contact_submit.

Conventions

en-GB spelling everywhere (“ageing”, “moisturiser”, “colour”).
Accessibility matters and has been actively fixed: keep the skip-link, :focus-visible rings, aria-expanded on nav/mobile/FAQ toggles, real <button>s for interactive controls, and WCAG-AA contrast (don’t reintroduce --text3 below ~4.5:1).
Privacy/consent: Google Analytics runs under Consent Mode v2, default-denied; the cookie banner grants analytics_storage. Don’t fire analytics before consent.
Assets: keep images right-sized (favicons are small PNGs; og-image.jpg is 1200×630 ~65KB). Don’t reintroduce multi-hundred-KB icons/OG images.
Self-contained: no external JS/CSS beyond Google Fonts (loaded non-render-blocking) and the analytics/newsletter/form third parties already present.

Protocol pages — the important workflow

Single-peptide protocol pages follow a fixed template: see _drafts/_PROTOCOL-TEMPLATE.md (front-matter, 13-section order, HTML skeleton, formatting rules). Read it before creating or editing any protocol page.

Key rules:

One page per vial size (e.g. ghk-cu-50mg.html, tirzepatide-30mg.html) — not one page per compound.
House dosing model (the site owner’s own protocol style, distinct from FDA/label): simple Low / Average / High dose bands; for metabolic peptides (tirzepatide/semaglutide/retatrutide) the weekly band is split across as many days as tolerated, aiming near-daily; titrate by results/side-effect tolerance. FDA/trial titration is kept only as cited context in the Research section, never as the primary how-to.
Signature format: an FAQ Q&A block (appetite/nausea, cancer, combining, route notes) and the Research: / Anecdotal: paragraph split. Label owner/community guidance [HOUSE PROTOCOL]/[ANECDOTAL], distinct from cited data.
Every dosing/efficacy claim is either a linked citation (rel="noopener noreferrer" target="_blank", PubMed/PMC/DOI) or explicitly labelled anecdotal.
Dosing tables always show mg + mL + U-100 units + the dose ÷ mg/mL formula. Front matter includes a calc: block powering the calculator deep-link.
Safety-first framing for prescription/GLP-1/investigational compounds; keep the “Research purposes only / not medical advice” callout. Investigational compounds (e.g. retatrutide) stay noindex.

Content workflow the owner wants: research (peptidedosages.com articles + the studies they link, deep-dived, + FDA labels/PubMed) → bullet-point draft in _drafts/ first for point-level review → expand to full per-size HTML only after sign-off. Reconcile cross-link slugs to real filenames at build time.

Working style

The owner wants throughput; parallel subagents have worked well for multi-peptide research/drafting.
Commit/push only what’s asked; drafts can be committed freely (they don’t publish). End commit messages with the Co-Authored-By trailer.
Files are LF; git warns about CRLF conversion on Windows — harmless.