← Back to Blogs
GH Release

OpenClaw v2026.5.16-beta.5 Release Summary

323937119 May 17, 2026

OpenClaw v2026.5.16-beta.5 Release Summary

OpenClaw v2026.5.16-beta.5 brings a comprehensive set of updates focused on user experience, system reliability, and expanded tool capabilities. This release is particularly significant for Mac app users and those utilizing the native Codex runtime, introducing a redesigned settings interface and critical fixes for session isolation and context management.

Key Changes

User Interface & Experience

  • Mac App Redesign: The Settings pages have been completely overhauled with consistent card layouts, cached navigation to prevent content blanking during tab switches, and a cleaner native sidebar. The Channels and Config settings now load significantly faster by deferring schema work and caching status rows.
  • CLI Enhancements: Added openclaw sessions list as an alias for openclaw sessions to maintain consistency across the CLI. PowerShell users now receive concrete profile paths and actionable reload commands during shell-completion setup.

Codex & Agent Runtimes

  • Context Budgeting: To prevent performance degradation in large sessions, the Codex app-server now rotates stale bindings when rollouts exceed 70k tokens and caps dynamic tool results before they enter the native thread.
  • Session Isolation: Fixed a critical issue where concurrent Codex turns for different agents could abort one another. Bound Codex sessions now preserve the owning agent directory, ensuring follow-up turns are routed through the correct agent runtime.
  • Provider Routing: Explicit OpenAI Codex harness selections are now correctly routed through the openai-codex provider, ensuring OAuth profiles are used instead of falling back to API keys.

Skills & Tooling

  • New Skills: Introduced a meme-maker skill (supporting SVG/PNG and Imgflip), a Python debugging skill (pdb, debugpy), and node inspector debugging tools.
  • Tool Policy Enforcement: Inline skill tool dispatch now strictly honors the effective tool policy (allow/deny/sandbox) before selecting a target.
  • Plugin SDK: Added defineToolPlugin and new CLI commands (build, validate, init) for typed simple tool plugins with generated manifest metadata.

Channel & Integration Updates

  • Telegram: Improved reliability for media groups; the bot now warns users if some photos in an album fail to download rather than silently dropping them. Reply-chain context is also preserved, allowing the bot to understand the history of a conversation when a user replies to a bot-generated image.
  • WhatsApp: The forceDocument flag is now honored end-to-end, allowing images and videos to be sent as uncompressed documents.
  • Proxy Support: Added first-class support for HTTPS managed forward-proxy endpoints and scoped CA trust via proxy.tls.caFile.

Impact

Stability & Performance

  • Subagent Reliability: Fixed a race condition where subagent completion announcements were marked as "announced" before they actually reached the parent transcript. Additionally, successful keep-mode completions that exhaust their retry budget are now suspended rather than terminally cleaned up, preventing data loss.
  • Memory Core: Improved startup performance and accuracy by adding a catch-up scan that compares on-disk session transcripts against the index, eliminating the need for --force reindexing after restarts.
  • CLI Performance: The openclaw models command is now up to 72% faster in plugin-heavy installations by reusing a metadata snapshot instead of repeatedly scanning the plugin registry.

Bug Fixes

  • Routing Fixes: Resolved a "route drift" issue where the message tool would incorrectly send replies to WebChat when the session key indicated a different channel (e.g., Signal).
  • Error Classification: Provider conversation-state rejections (such as role-ordering failures) are now classified and reported as clear user messages instead of generic runner failures.
  • Android Security: Added a confirmation prompt for Gateway TLS thumbprint changes, allowing users to accept certificate rotations instead of facing a hard connection failure.
  • Billing Accuracy: Moonshot/Kimi 429 errors are now correctly classified as billing issues rather than generic rate limits.

References

Pull Requests

Releases