The Kit Gets Its Own Front Door

Why Care?

The kit's README told developers what it was, but not how to start or why the conventions are shaped the way they are. Now there's a site for that, at lossless-group.github.io/context-vigilance-kit , in the same design as our corpus site: one brand, two sites. The corpus shows our own work at scale. This one is for the people adopting the kit.

What's New?

• Getting Started.

  • Install in Claude Code with two commands. Other agents copy the skills in; anything that reads AGENTS.md gets a snippet.

  • A first cycle, one command at a time, with the role each command gives the agent.

  • What lands in your repo.

  • The recommended companions (Graphify, Archify, the Chroma skills), with their own install commands.

• Twenty patterns, each saying what it is, why it exists, and how to apply it:

  • Canonical (eleven): folders and modes, frontmatter as contract, IDs minted by command, three date families, honest status, lede discipline, the prep arc, fork early, three readers, the drift policy, four-part versions.

  • Experimental (five): handoffs, clarified loops, decisions, habits, contracts.

  • Emerging (four): freedom labels, checklists with go-back lines, roles bound in config, hooks for exact rules only.

  All are written for the site; none are copied from real project docs.

• The kit's own changelog and context-v/ , rendered and searchable. Docs marked publish: false stay off the site.

• Everything the corpus splash had:

  • DESIGN.md

  • the three-mode theme (dark, light, vibrant)

  • Pagefind search with kind: and tag: filters

  • sort controls

  • the OG share images

  • llms.txt and llms-full.txt

  • the sitemap and robots.txt

  • production-only analytics

Update, same day: rendered through LFM. Every doc on the site is now parsed with @lossless-group/lfm (0.6.0), the Lossless Flavored Markdown package, using the same components as our other splashes. Callouts, hex-code citations, tables, and code blocks follow LFM. Wikilinks resolve to the linked page on this site, or to the file on GitHub; links into other repos stay as plain text. Dates also display correctly now: they were showing a day early in time zones west of UTC.

Update, same day: Vercel-ready. The same build now also deploys on Vercel from the domain root. astro.config.mjs switches its base path when it sees Vercel's build environment (or a SITE_URL ), robots.txt became an endpoint so its sitemap URL is right on either host, and splash/vercel.json holds the build settings. The splash README covers the one-time Vercel setup and OpenPanel analytics on both hosts.

Known Gotchas

• Pages has to be on. The workflow deploys on push to master and tries to enable Pages itself. If the first run fails, set Pages to "GitHub Actions" in the repo settings.

• Analytics need a client ID on each host. Without OPENPANEL_CLIENT_ID (a repo Variable for Pages, an environment variable on Vercel), the site runs with no analytics. The OpenPanel client must also list each origin under supported domains, or events are rejected.

What's Next?

• INSTALL.md and DEPENDENCIES.md in the repo, which Getting Started can then link to.