OpenClaw: Codex Fix Reverted in 2026.5.6 + 30+ Improvements

OpenClaw 2026.5.6 update

🚀 OpenClaw 2026.5.6 update: 5.5 Codex fix reverted + multiple improvements

OpenClaw on May 6 released 2026.5.6 version, fixing an important issue in 2026.5.5, and also shipped 5.5, a major release that includes 30+ fixes.

📅 2026.5.6 release (May 6)

Key fix: reverting the 5.5 Codex OAuth routing fix

The problem: the doctor –fix change in 2026.5.5 rewrote valid openai-codex/* ChatGPT/Codex OAuth routing into openai/*, which could cause:

OAuth-only GPT-5.5 setups failing
Unexpectedly switching to OpenAI API-key routing

The fix: 2026.5.6 reverts that change and restores the correct routing.

How to recover

If 2026.5.5 already changed your default model:

# 恢复 Codex OAuth 路由
openclaw models set openai-codex/gpt-5.5
openclaw config validate

Other fixes

Fix	Description
Plugins/runtime fetch	Strip third-party symbol metadata so valid plugin requests stop being rejected
Debug proxy	Normalize captured fetch header dicts so replayed requests don't fail
Web fetch	Limit guarded dispatcher cleanup so timed-out fetches don't leave live tool lanes behind

📅 2026.5.5 release (May 6) - major release

2026.5.5 is a major release with 30+ fixes:

Feishu fixes

Topic starter thread IDs: backfill missing native topic starter thread IDs so session routing works correctly
Session routing: first and follow-up turns stay in the same topic session

LINE fixes

dmPolicy validation: reject dmPolicy: “open” configs without a wildcard allowFrom
Webhook DMs: fail validation instead of silently blocking

Telegram/Codex fixes

Progress drafts: keep message-tool-only progress drafts visible
Native Codex tool progress: render each tool only once

xAI fixes

Grok 4.3 reasoning: stop sending OpenAI-style reasoning effort controls
Thinking profile: clamp to off to avoid sending unsupported reasoning levels

Discord fixes

Heartbeat ACK timeouts: measure from the actual heartbeat send to avoid fake reconnect loops
Guild routing: plain text control commands go through normal authorization and the mention gate
Streaming: show live reasoning text in progress drafts

Matrix fixes

Approval delivery: retry up to 3 times with brief backoff so transient send failures don't strand pending approvals

Slack fixes

Socket Mode SDK errors: preserve error context and structured API fields
Reconnect logs: startup failures no longer collapse into a bare unknown error

iOS pairing fixes

LAN/.local gateways: allow setup-code and manual ws:// connects
Tailscale/public routes: keep wss://
Mixed-auth reconnects: prefer explicit gateway passwords over stale bootstrap tokens

Control UI fixes

Compaction disclosure: compact N Checkpoint(s) disclosure
Session-level details: modern checkpoint history cards
Performance: keep chat and channel tabs responsive
Sessions: /new command and lifecycle hooks apply only to explicit Control UI session creation
Chat history: keep persisted assistant progress text visible
Runtime labels: show each session's agent runtime

TUI fixes

CLI respawn wrapper: skip interactive launches
Terminal loss: exit cleanly
Heartbeat sessions: refuse to resume as a remembered chat session
Session picker: limit to recent rows

Doctor fixes

Codex routes: repair legacy openai-codex/* routes
Heartbeat-poisoned: move to recovery keys
Gateway token: warn when OPENCLAW_GATEWAY_TOKEN shadows a different active source
Supervisor restarts: report in openclaw doctor –deep

Gateway fixes

Shutdown: cancel delayed post-ready maintenance
Maintenance/cron: suppress startup after quick restarts
HTTP: avoid loading managed outgoing-image media handlers
OpenAI-compatible: send assistant role SSE chunks
Status: avoid fast repeated samples being flagged as event-loop degraded
Supervisor restarts: show in openclaw gateway status –deep

Plugins fixes

Diagnostics: make source-only TypeScript package warnings actionable
Update: keep installed official npm and ClawHub plugins synced
Install: reassert managed npm plugin peer links

Providers fixes

Fireworks/Kimi: thinking-off-only to avoid rejected reasoning* parameters

Video/Audio fixes

Video generation: accept provider-specific aspect-ratio and resolution hints
720P → 768P: normalize to MiniMax's supported value
Google generateAudio: stop sending on Gemini video requests

WhatsApp fixes

Responsiveness: stop only verified stale local TUI clients

Agents fixes

Context engines: keep hidden runtime-context messages out
Generated media: treat attachment-style message tool actions as completed
Config: remove ambiguous legacy main agent dir helper

Exec fixes

Approvals: fall back to guarded copy when Windows rejects rename-overwrite

Hooks fixes

Session-memory: add collision suffixes to fallback filenames

CLI fixes

Status: show selected agent runtime/harness
Sessions: prune old unreferenced artifacts

🙏 Contributor thanks

This update thanks the following 20+ contributors:

@shakkernerd – Plugins/runtime fetch, Doctor/gateway, Gateway/status
@obviyus – Web fetch
@joeyzenghuan – Feishu
@Patrick-Erichsen – Matrix/approvals
@bryce-d-greybeeld – Discord/gateway
@NikolaFC – Discord/gateway
@ramitrkar-hash – Discord/guilds
@BunsDev – Control UI (multiple contributions)
@Alex-Alaniz – Exec approvals
@MilleniumGenAI – Exec approvals
@draix – iOS pairing
@googlerest – Plugins/diagnostics
@vincentkoc – TUI, Doctor, Gateway, Agents, Plugins (multiple contributions)
@frankekn – WebChat/Codex media, Providers/Fireworks
@yelog – Doctor/status
@slideshow-dingo – CLI/sessions

📝 Wrap-up

OpenClaw 2026.5.5/5.6 double release:

5.6: reverts the 5.5 Codex OAuth routing fix, avoiding GPT-5.5 OAuth setup failures
5.5: 30+ fixes covering Feishu, LINE, Telegram, Discord, Slack, iOS, Control UI, TUI, Doctor, Gateway, Plugins and more
20+ contributors involved
Performance and stability improvements across multiple components

OpenClaw keeps iterating and gets better every day.

GitHub Releases:https://github.com/openclaw/openclaw/releases

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