Summary

In the bundled get-reply-eY9NJdyX.js (source src/auto-reply/.../get-reply.ts), allowEmptyAssistantReplyAsSilent is gated behind isGroupChat &&, which makes it impossible to honor silentReplyPolicy: "allow" on direct (DM) chats. Empty/silent-token replies on direct chats always trigger the empty-response retry → fallback chain, even when configured to be allowed.

Repro

openclaw 2026.4.26. With:

"agents": {
  "defaults": {
    "silentReply": { "direct": "allow" }
  }
}

Send a heartbeat-style ping to a direct DM where the model would correctly respond with NO_REPLY. Observed: harness logs empty response detected → retrying with visible-answer continuation → empty response retries exhausted → incomplete turn detected stopReason=stop payloads=0 → surfacing error to user. Expected: silent absorption, since policy is "allow".

Code reference

plugin-runtime-deps/openclaw-2026.4.26-*/dist/get-reply-eY9NJdyX.js line ~2029:

const allowEmptyAssistantReplyAsSilent = isGroupChat && resolveGroupSilentReplyBehavior({
  sessionEntry,
  defaultActivation,
  silentReplyPolicy: silentReplySettings.policy,
  silentReplyRewrite: silentReplySettings.rewrite
}).allowEmptyAssistantReplyAsSilent;

The isGroupChat && short-circuit means direct chats always get false regardless of policy.

Suggested fix

Drop the isGroupChat && guard. The resolveGroupSilentReplyBehavior call's return value already encodes silentReplyPolicy === "allow" correctly (see silent-reply-Btab7kqB.js:618). The function name is a misnomer — it works for any conversation type once you let it run.

const allowEmptyAssistantReplyAsSilent = resolveGroupSilentReplyBehavior({...}).allowEmptyAssistantReplyAsSilent;

(Confirmed locally — patched the bundled JS, silentReply policy is now respected on direct chats.)

Related issues

May share root cause / overlap with:

• [Bug]: Telegram transcript contains final reply but no outbound send occurs for [thinking,text] turn on 2026.4.11 #66459 — Telegram transcript contains final reply but no outbound send occurs for [thinking,text] turn
• Gateway completeness check false-negative on pure-relay agents #72541 — Gateway completeness check false-negative on pure-relay agents
• Empty-response fallback leaks provider validation errors to users (self-propagating loop) #73020 — Empty-response fallback leaks provider validation errors to users (closed; possibly regressed)
• Support reasoning-field outputs and visible final-answer handling for native reasoning models #74021 — Support reasoning-field outputs and visible final-answer handling