OpenClaw 2026.6.6-beta.2 landed early Friday morning as a follow-up to the security-heavy 2026.6.6 beta train. The new prerelease was published at 03:32 UTC on June 12 and now includes full release verification, a dependency evidence bundle, and several channel-facing fixes that arrived after the first beta.
The broad release theme remains the same: tighter boundaries around agents, transports, channels, browser output, MCP, and provider sessions. But beta 2 is not just a republished changelog. It folds in fresh work around Telegram resilience, Docker packaging, provider thinking profiles, startup performance, and release proof.
What Beta 2 Adds
The release notes still lead with security boundaries across transcripts, sandbox binds, host environment inheritance, MCP stdio, Codex HTTP access, native search policy, elevated sender checks, deleted-agent ACP bypasses, loopback tools, Discord moderation, Teams group actions, and exec approval timeouts.
That matters because OpenClaw sits at the junction of local machines, chat surfaces, browser sessions, and automation tools. A boundary bug in one place can become a data leak, a confusing approval path, or an unsafe tool call somewhere else.
The new beta 2 tag also adds evidence that the package made it through the release pipeline. The GitHub release links to the npm package, registry tarball, dependency evidence zip, release SHA, validation runs, npm preflight, plugin publish, ClawHub publish dispatch, OpenClaw npm publish, Telegram beta E2E, and postpublish package acceptance.
For a beta, that proof is useful. It gives operators a better trail for deciding whether to test the release in a real environment.
Telegram Recovery Gets More Practical
Two post-beta-1 Telegram fixes are especially operator-facing.
PR #91997 fixes isolated polling behavior when Telegram returns a 409 Conflict from getUpdates. That error usually means another gateway, script, or poller is using the same bot token, or a stale webhook is still attached.
Before the fix, isolated ingress could treat that conflict like a fatal account crash. The merged change preserves the Telegram Bot API error code, restarts the isolated polling cycle, clears the webhook again, rotates transport, backs off, and surfaces a clearer duplicate-poller hint in channel status.
PR #92004 makes Telegram streaming previews less fragile. Instead of permanently killing a draft preview on any edit failure, OpenClaw now distinguishes harmless "message is not modified" responses, flood-control retry windows, retryable edit errors, and ambiguous send failures.
The practical result is a more durable live-preview path. A transient Telegram edit problem should no longer turn off streaming previews for the rest of a run.
Provider and Runtime Fixes
Beta 2 also includes fixes outside Telegram.
PR #92053 applies Claude thinking profiles to custom providers that route through the native anthropic-messages adapter. The bug it fixes was subtle: a renamed Anthropic-compatible provider could silently clamp --thinking xhigh to off because the thinking policy was keyed too narrowly.
PR #92087 bundles the QA Lab runtime into the Docker image, closing a packaging gap for deployments that rely on containerized validation. The release also lists a /models derived-registry rescan fix under performance, which should reduce unnecessary registry work in startup or model-listing flows.
Should You Try It?
This is still a prerelease, so most production users should treat it as an evaluation build. But if you run Telegram ingress, test Docker-based deployments, operate custom Anthropic-compatible provider routes, or care about the release proof behind each beta, 2026.6.6-beta.2 is worth a close look.
The full changelog and verification links are available on the OpenClaw 2026.6.6-beta.2 GitHub release.