Rewrite all four docker-compose pages as reference wrappers (Rosetta pattern)

[dkijania]

Resolves #1200. Blocked on MinaProtocol/mina#18827 until canonical recipes land on develop.

Summary

• Rewrites all four node-operator docker-compose.mdx pages as thin narrative wrappers that link to canonical recipes in MinaProtocol/mina, mirroring the existing pattern in docs/exchange-operators/rosetta/docker-compose.mdx.
• Removes ~600 lines of inline docker-compose.yml YAML across the four pages — image tags, image names, and daemon flags now live in one place (the mina repo) instead of being copy/pasted into mdx and drifting every release.
• Adds the two pages that were missing from the earlier scope (snark-workers, seed-peers) — original PR only covered block-producer and archive-node.

Pages rewritten

Page	Now points at
docs/node-operators/block-producer-node/docker-compose.mdx	mina/src/app/cli/docker-compose/block-producer/
docs/node-operators/snark-workers/docker-compose.mdx	mina/src/app/cli/docker-compose/snark-worker/
docs/node-operators/seed-peers/docker-compose.mdx	mina/src/app/cli/docker-compose/seed-peer/
docs/node-operators/archive-node/docker-compose.mdx	mina/src/app/archive/docker-compose/

Plus regenerated static/llms-full.txt.

Section structure (consistent across all four pages)

Mirrors docs/exchange-operators/rosetta/docker-compose.mdx:

1. Intro paragraph linking to the canonical dir in mina
2. Quick start (git clone + cp env + vi .env + docker compose up -d / make)
3. Services overview (table)
4. Configuration tables (Docker images / Network / Ports / role-specific tunables)
5. Data persistence
6. Make targets
7. Verifying the deployment
8. Clean start (with explicit warnings about key loss)

Why reference, not embed

The Rosetta page proves the pattern: a dense scrapeable wrapper that describes the contract (services, env vars, ports, make targets) while the actual docker-compose.yml — the part that drifts every release — lives in the mina repo as a single source of truth. Modern coding agents (Claude Code, Cursor) follow GitHub links and clone repos; they get a runnable workflow rather than YAML to paste, and they always see current tags. Naive RAG indexers are already served by static/llms-full.txt at the bundle level, so embedding YAML in mdx wasn't doing anything for them either.

See #1195 for the broader discoverability context this fits into.

Test plan

• Block producer page renders cleanly on the Vercel preview.
• Archive node page renders cleanly on the Vercel preview.
• SNARK workers page renders cleanly on the Vercel preview.
• Seed peers page renders cleanly on the Vercel preview.
• All four mina/... URLs resolve once Add Docker Compose deployment recipes for block-producer, snark-worker, seed-peer, and archive mina#18827 merges to develop.
• check-llms-txt CI check passes (regenerated and committed).
• validate-docker-images passes (the four touched pages no longer have inline tags; pre-existing tags in node-operators/data-and-history/rosetta.mdx and exchange-operators/rosetta/run-with-docker.mdx are unchanged).

Follow-up

Optional drift-detection CI in a separate PR: assert every github.com/MinaProtocol/mina/... URL in these pages resolves on the referenced branch, and that env-var / make-target tables match the canonical files. Mentioned in #1200.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
docs2	Ready	Preview, Comment	May 6, 2026 3:41pm

[glyh]

BTW, I'm curious on the value of actually running a seed node. Is it discoverable by new nodes joining in the network? AFAIK, we only use the seeds in the peer list url as "verified" seeds at the moment.