Yesterday afternoon, old.openclaw911.com returned HTTP 502 on both its homepage and its REST endpoint, while two control sites on the same network remained HTTP 200. This morning, the same site came back with HTTP 200 on both checks. The lesson is not just that the outage ended—it is that the diagnosis was possible because we
How to Tell a Site-Specific Outage from a General Network Failure Read More »
A working preview on the machine that produced it is not yet a reviewable deliverable. A recent workflow correction made that plain: local addresses, control-panel portals, temporary tunnels, and third-party build previews can all look convincing to the person who made the work while failing the actual reviewer. The test has to begin where the
A Review Link Is Not Deliverable Until It Opens Outside Your Workspace Read More »
A video can contain a valid audio track and still appear silent to the person watching the page. That was the real operational lesson from a recent production fix: the first repair was technically sound, but the public page continued to point to the earlier video asset. The problem was not the audio mix The
When Audio Sounds Missing, Start With the Exact Asset the Browser Loaded Read More »
When a security scan finds a credible threat, speed matters. But deleting first and asking questions later can turn a recoverable incident into an unexplainable one. In a recent OpenClaw operations response, suspicious website files were isolated outside the public web path rather than permanently removed. The response preserved hashes and recovery copies, then checked
Containment Is Not Cleanup: Preserve Evidence and Prove Recovery First Read More »
A familiar name in a dashboard is not enough evidence that an automated post will reach the intended destination. During a real OpenClaw publishing check, an account contained a connected profile named OpenClaw911. The scheduled workflow, however, required a different specific profile ID. The visible name made the mismatch easy to miss: one record looked
Why an Automation Destination Must Be Verified by ID, Not Its Name Read More »
A useful integration is not automatically a safe integration just because a provider has been chosen. In a recent OpenClaw implementation, a licensed property-search provider was selected as the preferred next source for live inventory. The work did not stop at that decision. The provider’s public terms showed that access, lead handling, renewal, and data
Why a Selected Data Source Should Stay Disabled Until Its Contract Is Real Read More »
A URL returning an error is not always proof that the workflow behind it is broken. In a real OpenClaw operations check, a signed webhook delivery reached its handler and returned the expected accepted response. But an ordinary browser-style GET request to the same address fell through to a different upstream route and returned an
A Webhook Health Check Must Test the Event, Not Just the URL Read More »
Connecting an AI assistant to everything at once is tempting. It is also how a useful tool becomes difficult to trust. In a recent OpenClaw build, we launched a private “second brain” with a deliberately narrow starting boundary: curated local Markdown files were searchable, while email, calendar, customer systems, and task platforms remained disconnected until
A Private AI Second Brain Needs Honest Boundaries Before It Needs More Connections Read More »
OpenClaw 2.0 is not one flashy feature. It changes how work is started, followed, moved, protected, and recovered. The changes that matter in real operation Do not overlook the two breaking migrations OpenProse is no longer bundled, and shipped codex/* and openai-codex/* routes move to openai/*. Back up configuration and state, test away from production,
OpenClaw 2.0: What Actually Changed? Read More »
