# Codex replies and final answers > Visible reply delivery, heartbeat turns, and bounded final-answer recovery - 网址：https://funcoding.ai/agents/openclaw/plugins/codex-harness-runtime/replies/ - 来源：OpenClaw 官方文档原文（英文），MIT 许可，同步于 2026-10-11 - 官方原文：https://docs.openclaw.ai/zh-CN/plugins/codex-harness-runtime/replies --- How final assistant text reaches the source conversation, and what happens when a Codex turn settles without one. Part of the [Codex harness runtime](https://funcoding.ai/agents/openclaw/plugins/codex-harness-runtime/) guide; [Where each section moved](https://funcoding.ai/agents/openclaw/plugins/codex-harness-runtime/#where-each-section-moved) lists every section. ## Visible replies and heartbeats Direct/source chat turns through the Codex harness default to automatic final assistant delivery for internal WebChat surfaces, matching the Pi harness contract: the agent replies normally and OpenClaw posts the final text to the source conversation. Set `messages.visibleReplies: "message_tool"` to keep final assistant text private unless the agent calls `message(action="send")`. Commentary follows the conversation's progress settings and may be hidden or temporary. When a question arrives while work continues, the agent is instructed to send its answer with `message(action="send", final=false)` and continue the task. The final response covers answers and results that have not already been delivered. This does not enable commentary or change streaming settings. Native Codex messages explicitly marked for asynchronous delivery use an independent delivery intent. They are delivered separately from optional progress and from the eventual final answer, including when block streaming is disabled. Codex heartbeat turns get `heartbeat_respond` in the searchable OpenClaw tool catalog by default so the agent can record whether the wake should stay quiet or notify. Heartbeat turns use the same Codex Default collaboration mode as ordinary chat turns. The heartbeat monitor's cron scratch is appended to the scheduled heartbeat user message when present. ## Attachments in a remote workspace With `appServer.remoteWorkspaceRoot`, final and asynchronous replies can use `MEDIA:./report.pdf` or an absolute path inside the remote workspace. Before releasing the app-server connection, the Codex plugin reads the attachment with bounded `command/exec` requests. The host's [reply-media capability](https://funcoding.ai/agents/openclaw/plugins/sdk-agent-harness/attempt-runtime/#reply-attachments-from-a-remote-workspace) applies read policy and stages the bytes for delivery. The original reply remains in the transcript; Gateway workspace copies are not used as a fallback. Files outside the remote workspace produce a labeled attachment failure. Copy them into the workspace before sending. HTTP references and managed `media://` attachments continue to work. ## Final answers after settled tool work For ordinary host-authenticated Codex turns that finish tool work without a visible answer, OpenClaw can request a bounded final-answer turn in a private temporary home. It uses the completed thread's model selection and the original host auth route or resolved profile, rather than selecting a model from outer request metadata. The existing environment, dynamic-tool, MCP, and native-hook restrictions remain. Completed actions are transcript evidence, not instructions to replay. Preserving a native model does not, by itself, disable host-authenticated finalization. Hidden background-task completion messages remain eligible for this recovery when they are part of the model's context. The host records their native turn identity on the admitted prompt while keeping them hidden in chat. Native transcript mirroring reuses that prompt as evidence; it cannot recreate an admitted prompt that disappeared. Messages explicitly excluded from model context remain outside native prompt annotation. Recovery reserves its existing limits for the complete current turn, then keeps the nearest whole earlier exchanges that fit. Older exchanges can be omitted, including a whole exchange that is too large. A notice identifies missing history when space permits; the finalizer is always instructed to state uncertainty about missing facts. Current evidence that exceeds the limits, invalid tool pairs, or unsupported content still makes recovery unavailable. Existing conversation history stays intact, and completed actions are never repeated. A Chat created through Codex Sessions is different: its private supervision connection owns native authentication. Stock Codex does not expose a generic tool-free summary operation that preserves that connection's account. OpenClaw marks this finalization context unavailable instead of choosing host credentials, copying native credentials, or starting another native turn. If a final reply is required, the host delivers its existing fallback: > The tool run finished, but no final summary was produced. I did not repeat any completed actions. The original completed outcome, native binding, and tool receipts remain intact. If the native turn failed and finalization cannot produce an answer, the reply instead explains the failure and the next step. For example, model capacity errors suggest waiting and retrying or choosing another model. The turn remains failed in chat and task progress; completed actions are not repeated. An explicit Stop remains a canceled turn. Native turns that return a final answer are delivered normally. The ordinary `homeScope: "user"` opt-in retains its documented private host-auth finalization; see [Auth and environment isolation](https://funcoding.ai/agents/openclaw/plugins/codex-harness-reference/#auth-and-environment-isolation).