OpenClaw Update: Enhanced Media Support, Agent Reliability, and DX Improvements

The latest set of merged pull requests for OpenClaw focuses on expanding integration capabilities, particularly in media processing, and hardening the reliability of agent orchestration. Key improvements target the developer experience (DX) and the stability of long-running agent tasks, ensuring that users receive clearer feedback during failures and more consistent environment handling.

Merged PRs

• Control UI/sessions: nest subagent sessions under parent with visual prefix (Original PR)
• feat(openrouter): add inbound audio STT support (Original PR)
• fix(agents): recover read offsets beyond EOF (Original PR)
• fix(agents): use run workspace for post-compaction context (Original PR)
• fix(active-memory): classify threaded Telegram DMs (Original PR)
• Fix iMessage health probe failure reporting (Original PR)
• fix(docker): avoid external Dockerfile frontend pull (Original PR)
• fix(doctor): render legacy migration panel as preview when --fix is not passed (Original PR)
• fix: surface explicit error on LLM idle timeout to prevent ghost turns (Original PR)
• Add message tool delivery hint to inbound context (Original PR)
• fix(onboarding): fix onboarding hangs instead of exiting (Original PR)

Key Changes

Media and Integration Enhancements

OpenClaw has expanded its media-understanding capabilities by adding inbound audio Speech-to-Text (STT) support for OpenRouter. Previously, the OpenRouter plugin only supported images; now, it can route transcription requests through OpenRouter's /audio/transcriptions endpoint. This is particularly beneficial for channels like Telegram, where voice notes can now be transcribed and fed into the agent prompt seamlessly.

Additionally, the Active Memory plugin now correctly classifies threaded Telegram Direct Messages (DMs). By stripping the :thread: suffix from session keys, the system can now recognize these as direct chats, enabling memory recall and transcript persistence for topic-based Telegram conversations.

Agent Orchestration and Reliability

Several fixes address "ghost turns" and context loss:

• LLM Idle Timeouts: A fix now ensures that if an agent hits an idle timeout after executing tools, an explicit error message is surfaced to the user. This prevents the turn from ending silently, providing a clear notification that the request timed out.
• Workspace Context: The post-compaction context refresh now correctly uses the queued run's workspaceDir instead of the process current working directory (process.cwd()). This ensures that critical startup files, such as AGENTS.md, are re-injected correctly even when the runner process is executing from a different directory (e.g., in CLI-backed follow-up turns).
• Read Tool Robustness: The agent read tool now gracefully handles offsets that exceed the end of the file (EOF), returning an empty range instead of crashing.

DX and Tooling Improvements

• iMessage Health Reporting: The openclaw status --deep command now correctly reports iMessage health failures when macOS denies Full Disk Access to the Messages database, replacing a misleading "OK" status with a canonical remediation string.
• Doctor Tool Refinement: The openclaw doctor command now renders proposed legacy migrations as a "preview" when the --fix flag is absent. This removes contradictory output where changes were listed in the past tense despite not being applied to disk.
• Docker Build Optimization: The root Dockerfile no longer forces an external pull of the Dockerfile frontend, reducing build timeouts and failures on WSL environments.
• Onboarding Fix: A hang in the openclaw onboard wizard was resolved by ensuring the Codex migration provider uses an isolated app-server client that closes immediately after detection.

Impact

These changes significantly reduce friction for both end-users and developers. The addition of OpenRouter STT expands the utility of OpenClaw in voice-heavy channels, while the reliability fixes in the agent orchestrator eliminate silent failures that previously left users guessing why a turn stopped.

From a maintenance perspective, the refinements to the doctor and onboarding tools ensure a smoother first-time setup and a more transparent configuration migration path. The fix for iMessage health probes is a critical improvement for macOS users, providing actionable guidance on system permissions rather than silent failures.