OpenClaw Development Digest: Windows Stability, Agent Reliability, and Channel Enhancements

OpenClaw's rapid development continues with a wave of recent merges that collectively enhance the platform's stability, performance, and user experience. This digest highlights key changes across several critical areas, from fundamental Windows compatibility to refined agent behaviors and more resilient channel integrations. These updates reflect a commitment to addressing user pain points and improving the overall developer and operational experience.

Merged PRs

• PR #77650 · fix(sessions): avoid per-row model resolution when selected model metadata is persisted
• PR #77287 · fix(replay-history): drop trailing stream-error placeholder before pr…
• PR #77602 · feat(cron): add agentId filtering to cron list
• PR #77705 · feat(status): show uptime in chat status
• PR #77686 · fix: recover Slack channel restart after stop timeout
• PR #77280 · fix(auth-profiles): exclude format rejections from profile cooldown
• PR #53788 · docs(windows): companion app copy and GitHub auth notes
• PR #74881 · fix(windows): resolve gcloud/gog/tailscale PATHEXT shims before spawn
• PR #77661 · fix(agents): rebuild sandbox skill prompts from sandbox workspace
• PR #77689 · fix(ui): hide heartbeat acknowledgements
• PR #77622 · Require real behavior proof for external PRs
• PR #42174 · fix: false error of Windows path when binding the host path to the sandbox.
• PR #77658 · fix(telegram): reuse preview for long text finals
• PR #77660 · fix: slack mention-gating thread participation
• PR #77544 · fix(plugins): repair missing openclaw peer links on update
• PR #77663 · fix(whatsapp): normalize onboarding allowlist numbers
• PR #76936 · Fix Codex bound thread recovery after restarts
• PR #74161 · fix(agents): enforce exact skill path from [AI-assisted]
• PR #77620 · fix: slack keep resumed sends in thread
• PR #69701 · fix(gateway): skip IPv6 ::1 loopback binding on Windows to prevent HTTP hang
• PR #73533 · fix(infra): skip POSIX /tmp preferred path on Windows (#60713)
• PR #77609 · fix: honor embedded runtime tool allowlists
• PR #77613 · docs(doctor): clarify configured plugin repair
• PR #77555 · feat(agents): post-compaction loop guard in pi-embedded-runner
• PR #75343 · fix(lsp): resolve Windows .cmd shims in spawnLspServerProcess
• PR #76593 · fix(media): use r+ instead of r for fs.open to fix EPERM on fsync on Windows
• PR #77565 · fix(doctor): clear stale Codex session state
• PR #77489 · fix(secrets): preserve auth profile key refs during provider scrub
• PR #77601 · [codex] Reuse compatible plugin metadata snapshots on hot catalog paths
• PR #77606 · Add instructions for how to setup slack for QA tests

Key Changes

This release window saw a strong focus on foundational stability and performance, alongside critical enhancements to agent intelligence and communication channels. Key change directions include:

Enhanced Windows Compatibility and Reliability

Several fixes directly address long-standing or newly discovered issues specific to Windows environments, making OpenClaw significantly more robust on this platform:

• Gateway Stability: A critical fix (#69701) prevents the gateway HTTP server from hanging on Windows by skipping IPv6 ::1 loopback binding, ensuring basic connectivity. Relatedly, the resolvePreferredOpenClawTmpDir function now correctly uses the Windows %TEMP% directory (#73533), resolving split state issues for log and temporary files.
• Tool and Plugin Execution: Problems with child_process.spawn on Windows, which struggled with .cmd shims, have been resolved. This specifically fixes gcloud ENOENT errors for Gmail webhooks (#74881) and enables LSP servers to start correctly (#75343).
• Sandbox and File Operations: Custom Docker bind mounts on Windows now correctly interpret drive letters (#42174), and fsync operations for media files no longer fail with EPERM (#76593), allowing image/attachment handling.

Agent Performance and Robustness

Improvements to how agents process information and interact with tools contribute to more reliable and efficient operations:

• Session Performance: listSessionsFromStore() now avoids expensive per-row model resolution for sessions without overrides, significantly speeding up openclaw sessions and Control UI loads, especially at scale (#77650).
• Loop Detection: A new post-compaction loop guard (#77555) prevents agents from getting stuck in repetitive tool-call loops immediately after context compaction, enhancing agent resilience.
• Skill Resolution: Sandboxed sessions now correctly rebuild skill prompt inputs from the sandbox workspace (#77661), preventing errors where agents tried to access unreadable host paths. Additionally, agents are now explicitly constrained to use exact skill paths from <available_skills>, mitigating path hallucination (#74161).
• Tool Access Control: The embedded runner now properly honors tool allowlists, ensuring agents only access permitted tools (#77609).

Channel Integration Enhancements

Several updates improve the reliability and user experience across various communication channels:

• Slack: Critical fixes ensure Slack channels recover correctly after event-loop starvation (#77686), prevent untagged replies from being ignored in active threads (#77660), and maintain thread context for resumed subagent sends (#77620).
• Telegram: Long text responses in Telegram now reuse the active streamed preview as the first chunk, eliminating temporary extra message bubbles and improving the flow of long replies (#77658).
• WhatsApp: Onboarding allowlist numbers are now normalized, ensuring consistent matching for sender IDs regardless of input format (#77663).

Core System and Developer Experience Improvements

General system hygiene, security, and developer tooling also received attention:

• Auth Profile Management: A significant fix prevents a single malformed request from triggering a provider-wide cooldown (#77280), ensuring other healthy sessions are not affected. Additionally, secrets apply now correctly preserves keyRef in auth profiles during provider scrubs, preventing silent credential loss (#77489).
• Plugin Management: Missing openclaw peer links for npm-installed plugins are now repaired during updates (#77544), resolving ERR_MODULE_NOT_FOUND issues. Plugin metadata snapshots are also reused more effectively on hot catalog paths, reducing latency and event-loop pressure (#77601).
• Diagnostics and UI: The /status command now includes gateway and host uptime (#77705), and HEARTBEAT_OK acknowledgements are hidden in the UI to reduce visual clutter (#77689).
• CLI Tools: The cron list command gains agentId filtering (#77602), allowing users to view jobs relevant to specific agents in multi-agent setups. The doctor command's documentation has been clarified for plugin repair (#77613), and it now clears stale Codex session state (#77565).
• Internal Quality: A new requirement for