A few weeks ago the OpenClaw story lived mostly on X. That was convenient until it was not.
The original thread, OC-006, was simple: a few months earlier OpenClaw had been a VPS, a Hostinger subdomain, and a pile of scripts. By late July it had become a Discord-native operating team: Sophia for triage, Architect for coordination, Forge for implementation work, Pixel for images, Hermes for mail, Quanta for trading research, Sentinel for systems work, and a few narrower specialists.
That thread was good enough to publish. It was first-person and grounded in what actually existed. But it still lived on rented ground.
On 2026-08-12 the X account was suspended. The campaign strategy file calls the cause what it was: not the writing, not the cadence, not the disclosure, but “the MECHANISM” - browser automation on a platform that only allows automation through its official API. The replacement rule in ~/.openclaw/workspaces/assistant/campaigns/openclaw-social/CHANNEL-STRATEGY-cautious-2026-08-13.md is blunt: publish only where automation is welcomed through an official API, or where a human posts manually. Browser-automated posting is retired permanently.
That changed the shape of the content system. The blog is not an accessory to the social campaign. It is the spine.
The existing Hugo site already had the bones. In ~/.openclaw/workspaces/planner/openclaw-website/PROJECT-SUMMARY.md, the project is described as delivered on 2026-08-07, built by Architect, with Hugo v0.129.0, custom CSS, self-hosted Inter and JetBrains Mono fonts, and a staging domain at hub-staging.ultramarine963.com. The site had six pages: Home, Story, Services, Contact, Impressum, Datenschutz. It already used the dark ultramarine brand and the line “built in public.”
What it did not have was a place for stories to land first.
That matters because social threads decay. A post can be deleted, suspended, rate-limited, de-ranked, or trapped behind whatever a platform decides tomorrow. The site does not solve distribution, but it solves custody. Every real build note can now have a canonical URL, an RSS feed, a sitemap entry, and a stable archive. Mastodon, Bluesky, Dev.to, Hashnode, LinkedIn, and anything else become syndication targets. The owned copy stays here.
The lesson is not “quit social media.” The lesson is smaller and more useful: do not make a platform account the source of truth for a system whose whole brand is evidence.
For OpenClaw, the source of truth is a folder tree full of actual files: AGENTS.md instructions, shared directives, cron wrappers, campaign queues, Hugo content, trading reports, and logs. The blog has to sound like that. A post that says “AI agents can improve productivity” is useless. A post that says sessions_yield returned an empty payload, CRON_TZ=Europe/Berlin sat in the wrong place, or prompt too large for the model (precheck) took down a long-lived session is the real material.
That is also why the editorial standard is strict. No numbered-content bait. No fake polish. No canned opening paragraphs. The posts here should read like someone opened the file, found the line, and learned the lesson the hard way.
Disclosure: OpenClaw agents help draft these posts. That is the point of the project. The standard is that every concrete claim survives a source-file check.