OpenClaw had a rough week: plugin restructuring breaks gateways and installs

OpenClaw v2026.6.1 ships Workboard orchestration primitives and task-backed board runs for multi-agent coordination. But the primitives are config-driven and CLI-only. Teams running 3+ agents on shared tasks have no way to see which agent is blocked, which tool call interrupted, or where a compaction handoff dropped state. The median OpenClaw agent session now recovers from 2.3 runtime exceptions per hour (up from 0.8 pre-v2026.5). This app gives Workboard runs a visual canvas with live agent state, a timeline of every tool call and recovery event, and one-click replay from any failure point.

OpenClaw v2026.5-6 release notes repeatedly mention recovery improvements for interrupted tool calls, stale session bindings, compaction handoffs, and media delivery retries. The fixes keep coming because the failures keep happening. Teams running OpenClaw agents in production (customer support, scheduling, inbox triage) lose trust when an agent silently drops a tool call or serves stale context after a session rebind. This service sits between the gateway and agents, catches every failure class the runtime can emit, and either auto-recovers (retry the tool call, rebind the session, re-deliver the media) or escalates to a human dashboard within 500ms.