The OpenClaw team published a short, direct blog post today acknowledging a rough stretch for the project — and laying out what they are doing about it. It is worth reading in full. Here is the breakdown.
What Went Wrong
The problems started around April 24th. By April 29th, with the v2026.4.29 release, they were obvious:
"Gateways got slower. Some installs got stuck in plugin dependency repair loops. Discord, Telegram, WhatsApp and other channels behaved worse than they should. People downgraded. People lost time."
The root cause was not a single bug. The team had been migrating OpenClaw from a monolithic bundled architecture toward a plugin-based model — moving channels, providers, heavy tools, and optional integrations out of core and onto ClawHub. The goal is a leaner, more security-auditable core.
The problem: they underestimated how hard the transition would be to get right.
"Plugin dependency repair ran in startup and update paths, bundled and external plugins were half-split, ClawHub artifact metadata was still settling, and gateway cold paths did too much work."
The result was a worst-case middle state: too much had moved toward plugins, while too many plugins were still bundled, repaired, staged, and dependency-loaded in places that users feel immediately on every gateway start and update.
The Supply Chain Context
The timing overlapped with broader npm ecosystem anxiety. The team notes that OpenClaw did not directly depend on the Axios supply chain compromise, but the incident made the shape of the dependency graph feel a lot less theoretical:
"Transitive packages, install-time behavior, postinstall scripts, packages pulling packages pulling packages."
This is part of why the migration toward a smaller core is happening — not just to make OpenClaw faster, but to reduce the attack surface.
What Changes
Smaller Core, Clearer Boundaries
The plugin inventory now documents what ships in core, what installs separately, and what is source-checkout only. The work continues, but with more care around transition states.
A Real Team
The post is refreshingly candid about an organizational problem:
"OpenClaw was still too founder-driven. Too much release, review, packaging and support work sat with me."
Through the OpenClaw Foundation, and with support from OpenAI, the team is building a proper engineering and support organization around the project. This is arguably the most significant announcement in the post.
LTS Coming in May
The team is changing how releases work and plans to announce a Long-Term Support (LTS) release track "later in May" alongside the existing fast update cycle. No further details yet, but this is a direct response to operators who need stability over features.
Where Things Stand Now
Today's stable v2026.5.4 release is in many ways the first post-rough-week release — the gateway performance improvements, the plugin update pipeline fixes, and the doctor tool hardening all trace back directly to the problems identified in the postmortem. The changelog is long, but a significant portion of it is cleanup of exactly the transition debt that caused the instability.
The HN thread on the blog post (story 48021298) and a parallel submission (48029335) generated moderate discussion. The community response has been generally sympathetic — most users appreciate the candor.
The full post is at openclaw.ai/blog/openclaw-rough-week.