[Release] openclaw/openclaw v2026.5.10-beta.1: openclaw 2026.5.10-beta.1

OpenClaw v2026.5.10-beta.1 introduces significant enhancements to QA automation, real-time voice diagnostics, and gateway security. Key additions include Telegram live PR evidence automation with Crabbox integration, Discord voice diagnostics, and private skill archive uploads. The release also includes numerous bug fixes across CLI, agent routing, model compatibility, and chat delivery systems, plus dependency updates for improved performance and security.

Key Points

• •Telegram QA automation: Added live PR evidence capture with Convex-leased credentials, Crabbox transcript capture, motion GIF previews, and inline PR comments for automated testing workflows
• •Discord voice diagnostics: Implemented real-time voice analysis including speaker turn detection, playback reset handling, barge-in detection, and audio cutoff analysis
• •Telegram Desktop scenario builder: New tool that leases Crabbox, installs native Telegram Desktop, configures OpenClaw gateway with bot credentials, and records VNC artifacts
• •Real-time voice instructions: Added `talk.realtime.instructions` to allow operators to append custom voice style guidance while preserving built-in agent-consult behavior
• •Private skill archive uploads: Gated opt-in feature (`skills.install.allowUploadedArchives`) enabling trusted Gateway clients to stage and install zip-backed skills with explicit operator approval
• •Pure-JS Opus decoder default: Switched to `opusscript` decoder by default to avoid slow native addon compilation outside dedicated voice-performance lanes
• •Agent session scoping fix: Preserved scoped background exec/process session references across embedded compaction without exposing unrelated scope sessions
• •CLI/onboarding improvements: Enhanced setup, configure, and channel commands with better wayfinding that explains next useful commands instead of terse labels
• •Model compatibility fixes: Resolved issues with OpenAI-compatible models returning JSON bodies, xAI Grok reasoning exposure, and Anthropic-compatible thinking block handling
• •Security and stability: Fixed silent model hangs with LLM idle watchdog, improved error handling for malformed requests, and enhanced console capture to prevent session key leakage

Workflow Diagram

Concepts

Artifacts (4)

skills:
  install:
    allowUploadedArchives: true  # Enable private skill archive uploads for trusted Gateway clients

talk:
  realtime:
    instructions: "<custom voice style instructions>"  # Append operator-defined voice guidance while preserving built-in agent-consult behavior

discord:
  voice:
    decoder: "opusscript"  # Pure-JS Opus decoder (default, avoids native addon compilation)
    diagnostics:
      enabled: true
      features:
        - speaker_turn_detection
        - playback_reset_handling
        - barge_in_detection
        - audio_cutoff_analysis
    interruptResponseOnInputAudio: false  # Honor disabled input-audio interruption locally

# Handle managed select button callbacks before raw callback fallback
# Preserve delimiter-containing option values such as 'env|prod'
# Fixes issue #79816

function handleTelegramCallback(callback) {
  // Process managed select button callbacks first
  if (isManagedSelectButton(callback)) {
    return processManagedSelect(callback);
  }
  // Fall back to raw callback handling
  return processRawCallback(callback);
}