OpenClaw v2026.5.5 and v2026.5.6: The Recovery Releases

OpenClaw shipped two releases today — v2026.5.5 (9 AM UTC) and v2026.5.6 (5:51 PM UTC) — and together they represent the most concentrated burst of bug fixes since the "rough week" post. Both releases went out on May 6th, and if you're still on v2026.5.4, there's a lot to catch up on.

Here's what landed.

v2026.5.5: The Big Cleanup

This release touches almost every corner of the stack. Highlights below.

Channel Fixes

Feishu now correctly hydrates missing native topic starter thread IDs before routing sessions, which fixes a long-standing issue where first turns and follow-ups would land in different topic threads (#78262).

LINE now properly validates dmPolicy: "open" configs — if allowFrom is missing a wildcard, the webhook DM fails validation instead of being silently dropped (#78316).

Discord got two important fixes:

• Heartbeat ACK timeouts are now measured from the actual heartbeat send time, preventing false reconnect loops on startup (#77668).
• Plain text control commands like /steer now route through the normal authorization and mention gate instead of being silently dropped before an agent session can see them (#78080).

Telegram/Codex no longer duplicates tool progress lines in progress drafts (#75641).

Matrix approvals now retry delivery up to 3 times with backoff, so transient send failures don't strand pending approval prompts (#78179).

iOS pairing now allows setup-code and manual ws:// connects for private LAN and .local gateways while keeping Tailscale and public routes on wss://. It also prefers explicit gateway passwords over stale bootstrap tokens in mixed-auth reconnects (#47887).

Provider Fixes

xAI/Grok received two important patches: OpenClaw no longer sends OpenAI-style reasoning effort controls to native Grok Responses models (so xai/grok-4.3 stops failing with "Invalid reasoning effort" in Docker/Gateway runs), and the bundled xAI thinking profile is now clamped to off to prevent unsupported reasoning levels from leaking through.

Fireworks/Kimi K2.5 and K2.6 models are now exposed as thinking-off-only, preventing rejected reasoning* parameters from being sent (#74289).

TUI and Doctor Improvements

The TUI got several fixes to address the heartbeat-poisoned session problem that tripped up many users during the rough week:

• TUI now skips the generic CLI respawn wrapper for interactive launches and exits cleanly on terminal loss.
• It refuses to restore heartbeat sessions as the remembered chat session, preventing stale heartbeat history and orphaned openclaw-tui processes.
• Session picker is now bounded to recent rows with exact-lookup refreshes, so dusty stores no longer cause TUI to hydrate weeks-old transcripts before becoming responsive.

Doctor gained new repair capabilities:

• doctor --fix now repairs heartbeat-poisoned main session store entries and clears stale TUI restore pointers.
• The Codex repair path now fixes legacy openai-codex/* routes across primary models, fallbacks, heartbeat/subagent/compaction overrides, hooks, and stale session pins.

Gateway Performance

The Gateway startup path was significantly lightened:

• Model-catalog test helpers, QR pairing helpers, and TypeBox memory-tool schema construction are now kept off hot startup import paths.
• Non-readiness sidecars are deferred until after the ready signal.
• jiti is no longer imported on native-loadable plugin startup paths.

Control UI Polish

Multiple Control UI improvements landed thanks to @BunsDev:

• Chat tabs stay responsive while history payloads and channel probes are slow.
• Consecutive duplicate messages (like heartbeat ACKs) are collapsed into a single bubble with a count.
• Agent runtime is now shown in the Sessions table with runtime filter support.
• The cron New Job sidebar is now collapsible to reclaim space.
• Chat controls stay on one row on desktop with improved responsive behavior.

v2026.5.6: The Hotfix

Released just hours after v2026.5.5, this patch reverts one of the doctor --fix changes from the previous release.

The problem: The v2026.5.5 doctor repair rewrote valid openai-codex/* ChatGPT/Codex OAuth routes to openai/*, which could break OAuth-only GPT-5.5 setups or accidentally move users onto the API-key route instead of the Codex OAuth route (#78407).

If v2026.5.5 already changed your default model, you can recover by running:

openclaw models set openai-codex/gpt-5.5
openclaw config validate

Full recovery docs: docs.openclaw.ai/providers/openai#check-and-recover-codex-oauth-routing

Two other fixes also landed in v2026.5.6:

• Plugin/runtime fetch no longer passes third-party symbol metadata from header dicts into native fetch or Headers, preventing SDK and guarded fetch rejections (#77846).
• Web fetch now bounds guarded dispatcher cleanup after request timeouts, so timed-out fetches return tool errors instead of leaving Gateway tool lanes stuck active (#78439).

What to Do

If you're on v2026.5.4, upgrade to v2026.5.6 directly — it includes all of v2026.5.5 plus the OAuth route fix. If you ran v2026.5.5 and use Codex OAuth, check your default model with openclaw models list and apply the recovery steps above if needed.

Given the pace of fixes, it's also a good moment to run openclaw doctor --deep and see what it surfaces.

Sources: v2026.5.5 release, v2026.5.6 release, Recovery docs