Multi-Site Management: One Toolkit, N Sites

A true scene first

Three sites after the cloning lesson. Does Monday’s 30-minute freshness loop run three times? This lesson folds “N sites” back into “one command”: register once, inspect everything.

Register and inspect

# Register (once per site)
npx anvilwiki-ops sites add anvil-wiki /path/to/anvil-wiki
npx anvilwiki-ops sites add forge-wiki /path/to/forge-wiki --url https://forge.example

# Inspect
npx anvilwiki-ops --all audit            # site-wide health, one report (one failure doesn't halt)
npx anvilwiki-ops --all metrics          # all sites' data
npx anvilwiki-ops --site forge-wiki metrics   # just one
npx anvilwiki-ops sites list             # what's registered
npx anvilwiki-ops sites remove <name>     # remove

The registry lives at ~/.config/anvil-ops/sites.toml and stores names and paths only — credentials never enter it: each site’s .env (GSC key, CF token) stays in its own repo; nothing cross-pollinates.

AI referral tracking

metrics ends with an AI referrals section: visits arriving from chatgpt.com, perplexity.ai, gemini.google.com, claude.ai and peers (counted via your Cloudflare token). GSC’s generative-AI report has no API, so insights also probes AI Overviews inclusion (experimental), and metrics --import-aio <csv> imports the CSV GSC’s UI exports.

Reading discipline: trends, not absolutes — AI browsers often strip referrers, so low numbers are a tooling limit, not your failure. It’s the first observable signal of “is my site being AI-cited”, pairing with the direct-answer and tables playbook from the learning manual.

The safety line: submit deliberately refuses —all

Read commands (doctor/metrics/audit/insights) take --all happily; the write command submit deliberately supports single sites only — bulk publishing is too dangerous to one-click. Every site’s launch decision deserves to be made in front of its own diff.

Three classic mistakes (made for you in advance)

Credentials pasted into sites.toml: the registry is a plaintext file — keys in there are keys taped to the door. Credentials live only in per-site .env.
Wishing submit had —all: it’s designed not to. The bulk-launch urge belongs to the launch-batch discipline, not a flag.
Treating AI referral absolutes as KPIs: referrer stripping keeps them naturally low; an upward trend is the good news.

Three words to know (just these)

Registry (sites.toml): the name-and-path list of your sites, managed via sites add/remove/list.
—all / —site: inspection scope switches — everything or one; read commands only.
AI referrals: visit-source stats from AI assistants — the observable signal of “being AI-cited”.

✅ Acceptance (all must hold)

☐ Site two is sites add-ed and visible in sites list
☐ --all audit has produced one site-wide report
☐ You know where credentials live (per-site .env) and what the registry holds (names and paths only)

The Development Manual closes here

The map (architecture) → categories and languages → feature switches → verify your changes → sync upstream → contribute back → AI ops → the content pipeline → multi-site management — you now hold maintainer-grade command of this template. Return to the Learning Manual’s weekly loop: most of those 30 minutes now run themselves. When trouble strikes, the learning manual’s appendices hold the red-line master list, the troubleshooting tree, and the glossary.