OpenClaw Digest: Enhancing Messaging Reliability and System Performance

Merged PRs

• fix: surface hook model rejection diagnostics Original PR
• refactor: stage external output writes through fs-safe Original PR
• fix(whatsapp): dedupe captioned MEDIA auto-replies Original PR
• fix(ui): label inherited thinking overrides Original PR
• fix(discord): smooth voice capture and add voice QA smoke Original PR
• perf(slack): reduce message hot-path overhead Original PR
• perf(core): trim reply queue churn Original PR
• fix(whatsapp): resolve outbound PN to LID via auth-dir forward mapping Original PR
• perf(telegram): reduce message context setup work Original PR
• telegram: avoid silent fallback after message tool send Original PR
• fix(agents): clean subagent fallback scaffolding Original PR
• Fix Control UI node exec approvals Original PR
• chore(deps): bump @openclaw/fs-safe pin to 3412e03 Original PR
• fix(plugins): forward setChannelRuntime from non-bundled external setup entries Original PR
• fix(subagents): honor archiveAfterMinutes for session-mode reaping Original PR
• fix: rotate sessionFile after daily reset Original PR
• test(telegram): lock draft finalization ordering Original PR
• fix(telegram): keep polling watchdog on getUpdates liveness Original PR

Key Changes

Messaging Integration Fixes

Significant effort was directed toward solving "silent failures" in third-party integrations. For WhatsApp, a critical bug was fixed where proactive messages to certain contacts created "ghost chats" (sender-only threads) because the system failed to resolve phone numbers to LIDs (Local IDs). Additionally, WhatsApp auto-replies were deduplicated to prevent sending the same image twice when captions were present.

Telegram received several stability and UX updates. A critical fix ensures that the polling watchdog now specifically monitors getUpdates liveness; previously, unrelated outbound API activity could mask a stalled inbound polling loop, leaving the bot silent for long periods. Other updates prevent unwanted "Nothing further to report" fallback messages after a successful message.send tool call and lock the ordering of streaming progress messages to prevent the final reply from appearing before the process logs.

Performance Optimizations

Performance tuning focused on reducing overhead in high-traffic "hot paths."

• Slack: Reduced message preparation allocations and cached native stream recipient lookups to lower RTT.
• Core: Trimmed reply queue churn by reducing avoidable array allocations and temporary helper objects in the auto-reply and agent execution paths.
• Telegram: Optimized message context setup by skipping non-forum topic-cache setup and deferring status reaction variants.

Agent and Gateway Orchestration

Improvements were made to the transparency and lifecycle of agents:

• Diagnostics: Hook-triggered runs that are rejected due to model allowlist restrictions now surface structured warnings in the Gateway logs, preventing the "HTTP 200 but no action" confusion.
• Subagents: The delivery contract for subagents was cleaned up to prevent raw internal runtime scaffolding (e.g., BEGIN_UNTRUSTED_CHILD_RESULT) from leaking into external chats during mediated announce failures.
• Session Management: Fixed a regression where daily session resets updated the session ID but failed to rotate the sessionFile, causing transcripts to grow indefinitely across days.

Security and Infrastructure

• Node Execution: Fixed a bug in the Control UI where node execution approvals failed with APPROVAL_CLIENT_MISMATCH for token-authenticated users without a paired device.
• Filesystem Safety: External output writes (e.g., Playwright screenshots, ffmpeg transcodes) are now staged through fs-safe to ensure atomic publication and boundary hardening.

Impact

These changes primarily impact operators and users of the WhatsApp and Telegram channels by eliminating silent delivery failures and improving the chronological flow of conversations. The performance optimizations in the core reply queue and Slack integration reduce latency and resource consumption during high-volume agent interactions.

For developers and maintainers, the improved diagnostic surfacing for hook model rejections and the fix for the WeChat plugin runtime initialization timeout significantly reduce the time required to debug integration failures. The rotation of session files ensures that long-lived agents maintain a healthy context window without unbounded on-disk growth.