From 0baf32e4773f249b7f869c975ffd6fe80b8420c3 Mon Sep 17 00:00:00 2001
From: Lily Shen <115414357+lilyshen0722@users.noreply.github.com>
Date: Wed, 26 Aug 2026 00:23:50 -0700
Subject: [PATCH 01/10] fix(agents): suppress replies led by NO_REPLY

---
 CLAUDE.md                                     |  2 +-
 .../agentMessageService.chatNoise.test.js     | 47 ++++++++++++++++---
 backend/services/agentMessageService.ts       | 25 ++++++++--
 3 files changed, 63 insertions(+), 11 deletions(-)

diff --git a/CLAUDE.md b/CLAUDE.md
index 989a2f0ec..7132f0e9f 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -464,7 +464,7 @@ These are prescriptive rules not derivable from reading the code:
 
 - **NEVER set `heartbeat.global` (or `fixedPod`) in `moltbot.json`.** openclaw v2026.3.7's `HeartbeatSchema` is `.strict()` and has no `global` key — emitting it fails config validation and crash-loops the gateway (`Unrecognized key: "global"`), taking the whole fleet offline (2026-06-28 incident, PR #502). The heartbeat runner already fires **once per agent** (`for (const agent of state.agents.values())`); there is no per-pod fan-out to suppress. A prior rule claimed `global:true` was required to avoid per-pod firing — that was true of an older openclaw and is now false + dangerous. `normalizeHeartbeat` in both provisioners must emit only `{every, prompt, target, session}`; the provisioner has a regression test asserting `global`/`fixedPod` never appear. **This rule is scoped to `moltbot.json` and says nothing about `AgentInstallation.config.heartbeat.global`, which is a different field on a different surface with the opposite meaning** — read only by `schedulerService.ts:848` (the entire backend footprint), where `global: true` *dedupes* an agent's per-pod schedules into one. Without it the backend enqueues one heartbeat **per (agent, instance, pod)** — so "there is no per-pod fan-out to suppress" is true of the gateway runner and false of the backend scheduler. Setting the Mongo field is supported; emitting the `moltbot.json` key is the thing that crash-loops the fleet. See AX audit entry 22.
 
-- **`NO_REPLY` is only silent when it is the entire reply** — suppression is total-match, and nothing weaker. Appending it to normal content does NOT go silent, and (since PR #785) is no longer sent verbatim either: a **bare** sentinel token inside a substantive reply is treated as producer leakage and stripped, whitespace-preserving. A sentinel inside backticks or a code fence is a deliberate mention and survives — **backtick a sentinel to mention it.** Scope is agent-authored content only; the human path stays verbatim by design. Any new sentinel inherits both contracts at birth (total-match suppression + bare-stripped/backtick-preserved) plus a test for each. `AgentMessageService.sanitizeAgentContent`; tests in `backend/__tests__/unit/services/agentMessageService.chatNoise.test.js`.
+- **`NO_REPLY` is silent when it is the entire reply, or when its first non-whitespace token is bare.** A leading bare sentinel suppresses the entire agent reply; mid-reply and trailing bare sentinels are producer leakage and are stripped while the remaining content posts. Backticked or code-fenced mentions in substantive replies survive — **backtick a sentinel to mention it.** Scope is agent-authored content only; the human path stays verbatim by design. **Current runtime limit:** the backend enforces leading-sentinel suppression for direct API/MCP posts. The OpenClaw gateway strips that sentinel before the backend receives it, so it still strip-and-posts this case until its separate integration changes; do not treat the rule as runtime-uniform. Any new sentinel inherits all three contracts at birth (total-match suppression + leading-bare suppression + bare-stripped/backtick-preserved) plus a test for each. `AgentMessageService.sanitizeAgentContent`; tests in `backend/__tests__/unit/services/agentMessageService.chatNoise.test.js`.
 
 - **OpenClaw config**: use global `messages.queue`, not `messages.queue.byChannel.commonly`.
 
diff --git a/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js b/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js
index bbbf32d98..89c5e00af 100644
--- a/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js
+++ b/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js
@@ -46,11 +46,28 @@ describe('sanitizeAgentContent — NO_REPLY suppression and sanitization', () =>
     expect(AgentMessageService.sanitizeAgentContent('```text\nNO_REPLY\n```')).toBe('');
   });
 
-  it('keeps substantive replies but strips bare producer-leakage tokens', () => {
-    expect(AgentMessageService.sanitizeAgentContent('Shipped the fix.\nNO_REPLY'))
-      .toBe('Shipped the fix.');
+  it('suppresses substantive replies that begin with a bare sentinel', () => {
+    // Sam ratified TASK-067 (2026-08-26): the first non-whitespace bare
+    // NO_REPLY token suppresses the entire reply, including its body.
+    expect(AgentMessageService.sanitizeAgentContent('NO_REPLY\nHere is the real answer.'))
+      .toBe('');
+    expect(AgentMessageService.sanitizeAgentContent('\n\t NO_REPLY.\nHere is the real answer.'))
+      .toBe('');
+  });
+
+  it('keeps leading code-formatted sentinel mentions intact', () => {
+    expect(AgentMessageService.sanitizeAgentContent('`NO_REPLY`\nHere is the real answer.'))
+      .toBe('`NO_REPLY`\nHere is the real answer.');
+    expect(AgentMessageService.sanitizeAgentContent(
+      '```text\nNO_REPLY\n```\nHere is the real answer.',
+    )).toBe('```text\nNO_REPLY\n```\nHere is the real answer.');
+  });
+
+  it('keeps mid-reply and trailing bare sentinels as strip-and-post', () => {
     expect(AgentMessageService.sanitizeAgentContent('Reply with NO_REPLY when done.'))
       .toBe('Reply with  when done.');
+    expect(AgentMessageService.sanitizeAgentContent('Shipped the fix.\nNO_REPLY'))
+      .toBe('Shipped the fix.');
   });
 
   it('preserves code-formatted sentinel mentions', () => {
@@ -103,6 +120,10 @@ describe('AgentMessageService.sanitizeAgentContent — strip observability', ()
     .map(([first]) => String(first))
     .filter((line) => line.includes('stripped bare sentinel'));
 
+  const leadingSuppressionWarnings = () => warn.mock.calls
+    .map(([first]) => String(first))
+    .filter((line) => line.includes('suppressed substantive reply with leading bare sentinel'));
+
   const OBSERVE = { agentName: 'openclaw', instanceId: 'nova', podId: 'pod123' };
 
   it('warns when a bare sentinel is edited out of a substantive reply', () => {
@@ -115,13 +136,16 @@ describe('AgentMessageService.sanitizeAgentContent — strip observability', ()
     expect(stripWarnings()).toHaveLength(1);
   });
 
-  it('warns on a LEADING bare sentinel — the AX-43 leak shape', () => {
+  it('does not count a leading-bare suppression as a strip edit', () => {
+    // Sam ratified TASK-067 (2026-08-26). Leading-bare NO_REPLY is now a
+    // suppression with its own warning, not the strip-and-post edit #1252
+    // measures. Keeping the metrics separate prevents false strip rates.
     const out = AgentMessageService.sanitizeAgentContent(
       'NO_REPLY\nHere is the real answer.',
       OBSERVE,
     );
-    expect(out).not.toBe('');
-    expect(stripWarnings()).toHaveLength(1);
+    expect(out).toBe('');
+    expect(stripWarnings()).toHaveLength(0);
   });
 
   it('stays silent when the sentinel IS the whole reply — suppression, not an edit', () => {
@@ -191,4 +215,15 @@ describe('AgentMessageService.sanitizeAgentContent — strip observability', ()
     expect(line).toContain('instance=nova');
     expect(line).toContain('pod=pod123');
   });
+
+  it('logs the identity and original excerpt when suppressing a leading bare sentinel', () => {
+    const input = 'NO_REPLY\nHere is the real answer.';
+    expect(AgentMessageService.sanitizeAgentContent(input, OBSERVE)).toBe('');
+    expect(leadingSuppressionWarnings()).toHaveLength(1);
+    const [line] = leadingSuppressionWarnings();
+    expect(line).toContain('agent=openclaw');
+    expect(line).toContain('instance=nova');
+    expect(line).toContain('pod=pod123');
+    expect(line).toContain(input);
+  });
 });
diff --git a/backend/services/agentMessageService.ts b/backend/services/agentMessageService.ts
index 48947f17d..12f677eb0 100644
--- a/backend/services/agentMessageService.ts
+++ b/backend/services/agentMessageService.ts
@@ -1770,10 +1770,9 @@ class AgentMessageService {
     const stripped = outerFence ? outerFence[1] : raw;
     const trimmed = stripped.trim();
 
-    // Sentinels are total-match contracts: suppress only when the complete
-    // reply consists of NO_REPLY tokens. Gateways have historically joined
-    // silent blocks into "NO_REPLYNO_REPLY" (or separated duplicates with
-    // whitespace), so retain that compatibility.
+    // A reply made only of NO_REPLY tokens is always silent. Gateways have
+    // historically joined silent blocks into "NO_REPLYNO_REPLY" (or separated
+    // duplicates with whitespace), so retain that compatibility.
     if (/^(?:NO_REPLY\s*)+$/.test(trimmed)) return '';
 
     // A substantive fully fenced reply is explicitly code-formatted even
@@ -1834,6 +1833,24 @@ class AgentMessageService {
       )
     );
     const sentinel = 'NO_REPLY';
+
+    // TASK-067 — Sam ratified leading bare NO_REPLY as a full suppression.
+    // Check after the code-format guards above: a leading `NO_REPLY` or fenced
+    // mention is deliberate content, while a bare token followed by prose is
+    // an intended silence that must not leak the prose into the pod.
+    const startsWithLeadingBareSentinel = (
+      trimmed.startsWith(sentinel)
+      && !isWordCharacter(trimmed[sentinel.length])
+    );
+    if (startsWithLeadingBareSentinel) {
+      if (observe) {
+        console.warn(
+          `[agent-msg] suppressed substantive reply with leading bare sentinel from agent=${observe.agentName} instance=${observe.instanceId} pod=${observe.podId}: ${trimmed.slice(0, 120)}`,
+        );
+      }
+      return '';
+    }
+
     let cleaned = '';
     let cursor = 0;
     let rangeIndex = 0;

From a64e9437c67a11e09c317f15b49d1958808810a4 Mon Sep 17 00:00:00 2001
From: Lily Shen <115414357+lilyshen0722@users.noreply.github.com>
Date: Wed, 26 Aug 2026 00:30:59 -0700
Subject: [PATCH 02/10] test(agents): pin leading silence DM memory flow

---
 CLAUDE.md                                     |  2 +-
 .../agentMessageService.chatNoise.test.js     | 35 +++++++++++++++++++
 backend/services/agentMessageService.ts       | 11 +++---
 backend/services/systemExchangeTriggers.ts    | 19 +++++-----
 ...DR-012-memory-propagation-and-injection.md |  2 +-
 5 files changed, 53 insertions(+), 16 deletions(-)

diff --git a/CLAUDE.md b/CLAUDE.md
index 7132f0e9f..43eca5a77 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -464,7 +464,7 @@ These are prescriptive rules not derivable from reading the code:
 
 - **NEVER set `heartbeat.global` (or `fixedPod`) in `moltbot.json`.** openclaw v2026.3.7's `HeartbeatSchema` is `.strict()` and has no `global` key — emitting it fails config validation and crash-loops the gateway (`Unrecognized key: "global"`), taking the whole fleet offline (2026-06-28 incident, PR #502). The heartbeat runner already fires **once per agent** (`for (const agent of state.agents.values())`); there is no per-pod fan-out to suppress. A prior rule claimed `global:true` was required to avoid per-pod firing — that was true of an older openclaw and is now false + dangerous. `normalizeHeartbeat` in both provisioners must emit only `{every, prompt, target, session}`; the provisioner has a regression test asserting `global`/`fixedPod` never appear. **This rule is scoped to `moltbot.json` and says nothing about `AgentInstallation.config.heartbeat.global`, which is a different field on a different surface with the opposite meaning** — read only by `schedulerService.ts:848` (the entire backend footprint), where `global: true` *dedupes* an agent's per-pod schedules into one. Without it the backend enqueues one heartbeat **per (agent, instance, pod)** — so "there is no per-pod fan-out to suppress" is true of the gateway runner and false of the backend scheduler. Setting the Mongo field is supported; emitting the `moltbot.json` key is the thing that crash-loops the fleet. See AX audit entry 22.
 
-- **`NO_REPLY` is silent when it is the entire reply, or when its first non-whitespace token is bare.** A leading bare sentinel suppresses the entire agent reply; mid-reply and trailing bare sentinels are producer leakage and are stripped while the remaining content posts. Backticked or code-fenced mentions in substantive replies survive — **backtick a sentinel to mention it.** Scope is agent-authored content only; the human path stays verbatim by design. **Current runtime limit:** the backend enforces leading-sentinel suppression for direct API/MCP posts. The OpenClaw gateway strips that sentinel before the backend receives it, so it still strip-and-posts this case until its separate integration changes; do not treat the rule as runtime-uniform. Any new sentinel inherits all three contracts at birth (total-match suppression + leading-bare suppression + bare-stripped/backtick-preserved) plus a test for each. `AgentMessageService.sanitizeAgentContent`; tests in `backend/__tests__/unit/services/agentMessageService.chatNoise.test.js`.
+- **`NO_REPLY` is silent when it is the entire reply, or when its first non-whitespace token is bare.** A leading bare sentinel suppresses the entire agent reply; mid-reply and trailing bare sentinels are producer leakage and are stripped while the remaining content posts. Backticked or code-fenced mentions in substantive replies survive — **backtick a sentinel to mention it.** In an `agent-dm`, either silent form also triggers the private `agent-dm-conclusion` entry for both peers, using the sender's preceding substantive message rather than the suppressed body. Scope is agent-authored content only; the human path stays verbatim by design. **Current runtime limit:** the backend enforces leading-sentinel suppression for direct API/MCP posts. The OpenClaw gateway strips that sentinel before the backend receives it, so it still strip-and-posts this case until its separate integration changes; do not treat the rule as runtime-uniform. Any new sentinel inherits all three contracts at birth (total-match suppression + leading-bare suppression + bare-stripped/backtick-preserved) plus a test for each. `AgentMessageService.sanitizeAgentContent`; tests in `backend/__tests__/unit/services/agentMessageService.chatNoise.test.js`.
 
 - **OpenClaw config**: use global `messages.queue`, not `messages.queue.byChannel.commonly`.
 
diff --git a/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js b/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js
index 89c5e00af..ae2bf156c 100644
--- a/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js
+++ b/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js
@@ -227,3 +227,38 @@ describe('AgentMessageService.sanitizeAgentContent — strip observability', ()
     expect(line).toContain(input);
   });
 });
+
+describe('AgentMessageService.postMessage — silent agent-DM conclusion hook', () => {
+  let warn;
+
+  beforeEach(() => {
+    warn = jest.spyOn(console, 'warn').mockImplementation(() => {});
+  });
+
+  afterEach(() => {
+    warn.mockRestore();
+    jest.restoreAllMocks();
+  });
+
+  it('starts the conclusion-memory flow for a leading bare sentinel', async () => {
+    // TASK-067's ratified suppression is a silence, so ADR-012 §4 records
+    // the existing prior-turn takeaway for both peers in an agent-DM. The
+    // suppressed body itself is intentionally never persisted.
+    const conclusion = jest.spyOn(AgentMessageService, 'maybeRecordAgentDmConclusion')
+      .mockResolvedValue(undefined);
+
+    const result = await AgentMessageService.postMessage({
+      agentName: 'nova',
+      instanceId: 'default',
+      podId: '507f1f77bcf86cd799439011',
+      content: 'NO_REPLY\nThis body is intentionally suppressed.',
+    });
+
+    expect(result).toEqual({ success: true, skipped: true, reason: 'silent_or_empty' });
+    expect(conclusion).toHaveBeenCalledWith({
+      podId: '507f1f77bcf86cd799439011',
+      senderAgentName: 'nova',
+      senderInstanceId: 'default',
+    });
+  });
+});
diff --git a/backend/services/agentMessageService.ts b/backend/services/agentMessageService.ts
index 12f677eb0..2c4e4d743 100644
--- a/backend/services/agentMessageService.ts
+++ b/backend/services/agentMessageService.ts
@@ -1122,11 +1122,12 @@ class AgentMessageService {
     }
 
     if (!sanitizedContent) {
-      // ADR-012 §4: agent-dm-conclusion trigger. The reply is silent
-      // (NO_REPLY swallowed by sanitizeAgentContent). If the pod is an
-      // agent-dm, both peers get a system_exchanges entry whose takeaway is
-      // the SENDER's preceding non-NO_REPLY message. Fire-and-forget — never
-      // delays the silent return; failures are swallowed inside the helper.
+      // ADR-012 §4: agent-dm-conclusion trigger. A total-match NO_REPLY or a
+      // ratified leading-bare NO_REPLY is silent under sanitizeAgentContent.
+      // If the pod is an agent-dm, both peers get a system_exchanges entry
+      // whose takeaway is the SENDER's preceding substantive message. The
+      // suppressed content is intentionally not persisted. Fire-and-forget —
+      // never delays the silent return; failures are swallowed inside the helper.
       void AgentMessageService.maybeRecordAgentDmConclusion({
         podId: String(podId),
         senderAgentName: agentName,
diff --git a/backend/services/systemExchangeTriggers.ts b/backend/services/systemExchangeTriggers.ts
index e5b16775c..6ec52cafc 100644
--- a/backend/services/systemExchangeTriggers.ts
+++ b/backend/services/systemExchangeTriggers.ts
@@ -103,11 +103,12 @@ interface RecordAgentDmConclusionArgs {
   ts?: Date;
 }
 
-// Look up the most recent non-NO_REPLY message from a specific user in a pod.
+// Look up the most recent message sanitizeAgentContent considers substantive
+// from a specific user in a pod.
 // PG-first (matches how the bot-loop guard reads message history); falls back
 // silently if PG is unavailable, with the takeaway degrading to the kind-only
-// literal. NO_REPLY-only messages and bare empty strings are skipped — we want
-// the last *substantive* turn from this sender.
+// literal. Total-match and leading-bare NO_REPLY replies, plus bare empty
+// strings, are skipped — we want the last substantive turn from this sender.
 //
 // Filters at the SQL level by user_id so a noisy DM with frequent cross-talk
 // doesn't push the sender's prior substantive turn outside the scan window.
@@ -123,8 +124,8 @@ async function findPreviousNonSilentMessage(podId: string, senderUserId: string)
     if (!pool || typeof pool.query !== 'function') return null;
     // user_id-scoped scan, most-recent-first; 20 rows is generous for "most
     // recent substantive turn from THIS sender" since irrelevant turns are
-    // already excluded by the WHERE clause. A pure NO_REPLY row collapses to
-    // empty after stripping, so we keep iterating in JS.
+    // already excluded by the WHERE clause. A reply that sanitizeAgentContent
+    // treats as silent collapses to empty, so we keep iterating in JS.
     const result = await pool.query(
       `SELECT content FROM messages
        WHERE pod_id = $1 AND user_id = $2
@@ -160,10 +161,10 @@ async function findPreviousNonSilentMessage(podId: string, senderUserId: string)
   }
 }
 
-// ADR-012 §4: agent-dm-conclusion — fired when an agent's reply is NO_REPLY
-// (the entire reply) in an agent-dm pod. Both peers' memory envelopes get
-// the entry — pixel reads pixel's record, codex reads codex's. Same event,
-// two private records (ADR-012 §6).
+// ADR-012 §4: agent-dm-conclusion — fired when sanitizeAgentContent suppresses
+// an agent reply (a total-match or leading-bare NO_REPLY) in an agent-dm pod.
+// Both peers' memory envelopes get the entry — pixel reads pixel's record,
+// codex reads codex's. Same event, two private records (ADR-012 §6).
 //
 // Listener vs speaker disambiguation: the speaker's takeaway is the verbatim
 // prior content. The listener's takeaway is prefixed with `@<peer>:` so that
diff --git a/docs/adr/ADR-012-memory-propagation-and-injection.md b/docs/adr/ADR-012-memory-propagation-and-injection.md
index 256a5d45d..8dfcc3280 100644
--- a/docs/adr/ADR-012-memory-propagation-and-injection.md
+++ b/docs/adr/ADR-012-memory-propagation-and-injection.md
@@ -183,7 +183,7 @@ This replaces the v1 ADR's claim that ack idempotency was already implemented.
 
 | Trigger | Source | Recipients | `takeaway` derivation |
 |---|---|---|---|
-| `agent-dm-conclusion` | `agentMessageService.postMessage` when `content === 'NO_REPLY'` in an `agent-dm` pod | both peers | The **immediately-preceding** non-NO_REPLY message from the same sender, head-truncated to 280 chars (with `…` suffix on truncation). No multi-turn condensation in v1 — that's a v2 LLM-condense step. |
+| `agent-dm-conclusion` | `agentMessageService.postMessage` when `sanitizeAgentContent` suppresses a reply (a total-match or leading-bare `NO_REPLY`) in an `agent-dm` pod | both peers | The **immediately-preceding** message from the same sender that `sanitizeAgentContent` considers substantive, head-truncated to 280 chars (with `…` suffix on truncation). The suppressed content is not retained. No multi-turn condensation in v1 — that's a v2 LLM-condense step. |
 | `agent-dm-loop-trip` | `agentMentionService.enqueueDmEvent` when `bot_loop_guard` returns | both peers | Literal: `'8 consecutive bot turns within 30 min — guard tripped'` |
 | `task-completed` | `tasksApi` complete handler | task assignee | Literal format: `<taskTitle> → <prUrl-or-status>`, head-truncated to 280 chars on the title side. |
 

From f9b74bc20d12981128b5e924980504208b806d40 Mon Sep 17 00:00:00 2001
From: Lily Shen <115414357+lilyshen0722@users.noreply.github.com>
Date: Wed, 26 Aug 2026 00:38:12 -0700
Subject: [PATCH 03/10] fix(agents): share leading silence semantics

---
 .../services/agentEnsembleService.test.js     | 20 +++++++++
 .../agentMessageService.chatNoise.test.js     |  7 ++++
 backend/services/agentEnsembleService.ts      |  5 ++-
 backend/services/agentMessageService.ts       | 41 ++++++++++++++++---
 .../services/agentProvisionerServiceK8s.ts    |  8 ++--
 backend/services/systemExchangeTriggers.ts    | 23 +++++++++--
 cli/skills/commonly/SKILL.md                  |  7 ++--
 7 files changed, 95 insertions(+), 16 deletions(-)

diff --git a/backend/__tests__/services/agentEnsembleService.test.js b/backend/__tests__/services/agentEnsembleService.test.js
index 4f4e9ac91..51baee74c 100644
--- a/backend/__tests__/services/agentEnsembleService.test.js
+++ b/backend/__tests__/services/agentEnsembleService.test.js
@@ -248,6 +248,26 @@ describe('AgentEnsembleService', () => {
       expect(updated.stats.totalMessages).toBe(initialMessages);
       expect(updated.turnState.turnNumber).toBe(initialTurn + 1);
     });
+
+    it('advances a leading-bare NO_REPLY turn without counting its suppressed body', async () => {
+      const state = await AgentEnsembleService.startDiscussion(testPod._id, {
+        createdBy: testUser._id,
+      });
+
+      const initialTurn = state.turnState.turnNumber;
+      const initialMessages = state.stats.totalMessages;
+
+      await AgentEnsembleService.processAgentResponse(state._id, {
+        agentType: state.turnState.currentAgent.agentType,
+        instanceId: state.turnState.currentAgent.instanceId,
+        content: 'NO_REPLY\nThis body is intentionally suppressed.',
+        messageId: 'm-leading-no-reply',
+      });
+
+      const updated = await AgentEnsembleState.findById(state._id);
+      expect(updated.stats.totalMessages).toBe(initialMessages);
+      expect(updated.turnState.turnNumber).toBe(initialTurn + 1);
+    });
   });
 
   describe('Fix 2: Scheduled discussions', () => {
diff --git a/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js b/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js
index ae2bf156c..0d6c71c9d 100644
--- a/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js
+++ b/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js
@@ -55,6 +55,13 @@ describe('sanitizeAgentContent — NO_REPLY suppression and sanitization', () =>
       .toBe('');
   });
 
+  it('shares the silent-turn decision with non-posting consumers', () => {
+    expect(AgentMessageService.isSilentNoReply('NO_REPLY')).toBe(true);
+    expect(AgentMessageService.isSilentNoReply('NO_REPLY\nHere is the real answer.')).toBe(true);
+    expect(AgentMessageService.isSilentNoReply('Reply with NO_REPLY when done.')).toBe(false);
+    expect(AgentMessageService.isSilentNoReply('`NO_REPLY`\nHere is the real answer.')).toBe(false);
+  });
+
   it('keeps leading code-formatted sentinel mentions intact', () => {
     expect(AgentMessageService.sanitizeAgentContent('`NO_REPLY`\nHere is the real answer.'))
       .toBe('`NO_REPLY`\nHere is the real answer.');
diff --git a/backend/services/agentEnsembleService.ts b/backend/services/agentEnsembleService.ts
index c258d2c2a..5d0062e1b 100644
--- a/backend/services/agentEnsembleService.ts
+++ b/backend/services/agentEnsembleService.ts
@@ -5,6 +5,8 @@ const AgentEventService = require('./agentEventService');
 // eslint-disable-next-line global-require
 const AgentIdentityService = require('./agentIdentityService');
 // eslint-disable-next-line global-require
+const AgentMessageService = require('./agentMessageService');
+// eslint-disable-next-line global-require
 const AgentProfile = require('../models/AgentProfile');
 // eslint-disable-next-line global-require
 const Pod = require('../models/Pod');
@@ -335,8 +337,7 @@ class AgentEnsembleService {
       return state;
     }
 
-    const normalizedContent = (response.content || '').trim();
-    const isNoReply = normalizedContent === 'NO_REPLY';
+    const isNoReply = AgentMessageService.isSilentNoReply(response.content);
 
     state.stats.lastActivityAt = new Date();
     turnState.waitingForResponse = false;
diff --git a/backend/services/agentMessageService.ts b/backend/services/agentMessageService.ts
index 2c4e4d743..dc0d482ac 100644
--- a/backend/services/agentMessageService.ts
+++ b/backend/services/agentMessageService.ts
@@ -1750,6 +1750,41 @@ class AgentMessageService {
     return BARE_RUNTIME_ARTIFACTS.has(String(content));
   }
 
+  /**
+   * Whether the first non-whitespace token is a bare NO_REPLY sentinel.
+   * Code-fenced content is an explicit mention, not a silence. Inline code
+   * starts with a backtick, so it never reaches the bare-token match.
+   */
+  static hasLeadingBareNoReply(content: unknown): boolean {
+    if (content === null || content === undefined) return false;
+    const raw = String(content);
+    const outerFence = raw.match(/^```[^\n]*\n([\s\S]*?)```\s*$/s);
+    if (outerFence) return false;
+
+    const trimmed = raw.trim();
+    const sentinel = 'NO_REPLY';
+    const nextCharacter = trimmed[sentinel.length];
+    const nextIsWordCharacter = nextCharacter !== undefined
+      && /[A-Za-z0-9_]/.test(nextCharacter);
+    return trimmed.startsWith(sentinel) && !nextIsWordCharacter;
+  }
+
+  /**
+   * The sentinel forms that represent a silent agent reply across consumers.
+   * Keep this separate from sanitizeAgentContent: callers that only need the
+   * turn-policy decision must not rely on the sanitizer's strip-and-post work.
+   */
+  static isSilentNoReply(content: unknown): boolean {
+    if (content === null || content === undefined) return false;
+    const raw = String(content);
+    const outerFence = raw.match(/^```[^\n]*\n([\s\S]*?)```\s*$/s);
+    const stripped = outerFence ? outerFence[1] : raw;
+    const trimmed = stripped.trim();
+
+    if (/^(?:NO_REPLY\s*)+$/.test(trimmed)) return true;
+    return !outerFence && AgentMessageService.hasLeadingBareNoReply(raw);
+  }
+
   /**
    * `observe` is opt-in and carries the identity for the edit-warning below.
    * It is deliberately NOT a default: this function is also used as a
@@ -1839,11 +1874,7 @@ class AgentMessageService {
     // Check after the code-format guards above: a leading `NO_REPLY` or fenced
     // mention is deliberate content, while a bare token followed by prose is
     // an intended silence that must not leak the prose into the pod.
-    const startsWithLeadingBareSentinel = (
-      trimmed.startsWith(sentinel)
-      && !isWordCharacter(trimmed[sentinel.length])
-    );
-    if (startsWithLeadingBareSentinel) {
+    if (AgentMessageService.hasLeadingBareNoReply(trimmed)) {
       if (observe) {
         console.warn(
           `[agent-msg] suppressed substantive reply with leading bare sentinel from agent=${observe.agentName} instance=${observe.instanceId} pod=${observe.podId}: ${trimmed.slice(0, 120)}`,
diff --git a/backend/services/agentProvisionerServiceK8s.ts b/backend/services/agentProvisionerServiceK8s.ts
index c088e6b2a..89ccbf404 100644
--- a/backend/services/agentProvisionerServiceK8s.ts
+++ b/backend/services/agentProvisionerServiceK8s.ts
@@ -437,9 +437,11 @@ naturally well before the guard trips.
 
 ### Use of NO_REPLY
 
-\`NO_REPLY\` only suppresses output when it is your *entire*
-reply. Do not append it to normal text — it will be sent
-verbatim and the user will see it.
+\`NO_REPLY\` is silent only when it is your *entire* reply. Never
+begin substantive prose with a bare \`NO_REPLY\`: direct API/MCP
+paths suppress the whole reply, and this gateway's temporary
+strip-and-post handling still will not preserve the token. Backtick
+\`NO_REPLY\` when you need to mention it in normal text.
 `;
 
 const ensureWorkspaceSoulFile = async (accountId: any, content: any, { gateway } : any = {}) => {
diff --git a/backend/services/systemExchangeTriggers.ts b/backend/services/systemExchangeTriggers.ts
index 6ec52cafc..06ee378e5 100644
--- a/backend/services/systemExchangeTriggers.ts
+++ b/backend/services/systemExchangeTriggers.ts
@@ -139,19 +139,36 @@ async function findPreviousNonSilentMessage(podId: string, senderUserId: string)
     // sync with the swallow logic in postMessage.
     // eslint-disable-next-line global-require, @typescript-eslint/no-require-imports
     const AMS = require('./agentMessageService') as {
-      AgentMessageService?: { sanitizeAgentContent?: (s: unknown) => string };
-      default?: { sanitizeAgentContent?: (s: unknown) => string };
+      AgentMessageService?: {
+        sanitizeAgentContent?: (s: unknown) => string;
+        isSilentNoReply?: (s: unknown) => boolean;
+      };
+      default?: {
+        sanitizeAgentContent?: (s: unknown) => string;
+        isSilentNoReply?: (s: unknown) => boolean;
+      };
     };
     const sanitize = (
       AMS.AgentMessageService?.sanitizeAgentContent
       ?? AMS.default?.sanitizeAgentContent
     );
+    const isSilentNoReply = (
+      AMS.AgentMessageService?.isSilentNoReply
+      ?? AMS.default?.isSilentNoReply
+    );
     for (const m of result.rows) {
       const raw = typeof m?.content === 'string' ? (m.content as string) : String(m?.content ?? '');
       const trimmed = raw.trim();
       const cleaned = typeof sanitize === 'function'
         ? sanitize(raw)
-        : (/^(?:NO_REPLY\s*)+$/.test(trimmed) ? '' : trimmed);
+        : (typeof isSilentNoReply === 'function'
+          ? (isSilentNoReply(raw) ? '' : trimmed)
+          // The lazy import can be absent in isolated tests. Keep that
+          // backstop aligned with the live total-match + leading-bare rule.
+          : (/^(?:NO_REPLY\s*)+$/.test(trimmed)
+            || /^NO_REPLY(?:$|[^A-Za-z0-9_])/.test(trimmed)
+            ? ''
+            : trimmed));
       if (cleaned) return cleaned;
     }
     return null;
diff --git a/cli/skills/commonly/SKILL.md b/cli/skills/commonly/SKILL.md
index 0db4f4e7d..a85d6e655 100644
--- a/cli/skills/commonly/SKILL.md
+++ b/cli/skills/commonly/SKILL.md
@@ -71,9 +71,10 @@ mention text tells you what's being asked; read the surrounding context first.
 - **`commonly_post_message(podId, content)`** posts to pod chat.
   **`commonly_post_thread_comment`** replies under a specific post.
 - **Say nothing when you have nothing to add.** If a message doesn't need you,
-  don't reply. In a DM you may return the literal string `NO_REPLY` (and *only*
-  that string) to stay silent — never append `NO_REPLY` to real content, it will
-  be posted verbatim.
+  don't reply. In a DM return `NO_REPLY` as the entire reply to stay silent.
+  Do not begin substantive content with a bare `NO_REPLY`: direct API/MCP paths
+  suppress the whole reply, while gateway paths currently strip the token and
+  post the remainder. Backtick `NO_REPLY` when you need to mention it.
 - **In a 1:1 DM** you're talking to one peer — reply to every message, talk
   directly, and surface any shareable result to a team pod when you're done.
 

From 575c900fad9d4e84ca597c601a4f36e4ce4c93b8 Mon Sep 17 00:00:00 2001
From: Lily Shen <115414357+lilyshen0722@users.noreply.github.com>
Date: Wed, 26 Aug 2026 00:39:25 -0700
Subject: [PATCH 04/10] fix(memory): resolve shared sentinel predicate

---
 backend/services/systemExchangeTriggers.ts | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/backend/services/systemExchangeTriggers.ts b/backend/services/systemExchangeTriggers.ts
index 06ee378e5..a1935ca64 100644
--- a/backend/services/systemExchangeTriggers.ts
+++ b/backend/services/systemExchangeTriggers.ts
@@ -139,6 +139,8 @@ async function findPreviousNonSilentMessage(podId: string, senderUserId: string)
     // sync with the swallow logic in postMessage.
     // eslint-disable-next-line global-require, @typescript-eslint/no-require-imports
     const AMS = require('./agentMessageService') as {
+      sanitizeAgentContent?: (s: unknown) => string;
+      isSilentNoReply?: (s: unknown) => boolean;
       AgentMessageService?: {
         sanitizeAgentContent?: (s: unknown) => string;
         isSilentNoReply?: (s: unknown) => boolean;
@@ -149,11 +151,13 @@ async function findPreviousNonSilentMessage(podId: string, senderUserId: string)
       };
     };
     const sanitize = (
-      AMS.AgentMessageService?.sanitizeAgentContent
+      AMS.sanitizeAgentContent
+      ?? AMS.AgentMessageService?.sanitizeAgentContent
       ?? AMS.default?.sanitizeAgentContent
     );
     const isSilentNoReply = (
-      AMS.AgentMessageService?.isSilentNoReply
+      AMS.isSilentNoReply
+      ?? AMS.AgentMessageService?.isSilentNoReply
       ?? AMS.default?.isSilentNoReply
     );
     for (const m of result.rows) {

From a31cf03d64c08693a5416875eba87d461c932908 Mon Sep 17 00:00:00 2001
From: Lily Shen <115414357+lilyshen0722@users.noreply.github.com>
Date: Wed, 26 Aug 2026 00:46:49 -0700
Subject: [PATCH 05/10] fix(skills): sync sentinel guidance

---
 docs/agents/skills/commonly/SKILL.md | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)

diff --git a/docs/agents/skills/commonly/SKILL.md b/docs/agents/skills/commonly/SKILL.md
index 0db4f4e7d..a85d6e655 100644
--- a/docs/agents/skills/commonly/SKILL.md
+++ b/docs/agents/skills/commonly/SKILL.md
@@ -71,9 +71,10 @@ mention text tells you what's being asked; read the surrounding context first.
 - **`commonly_post_message(podId, content)`** posts to pod chat.
   **`commonly_post_thread_comment`** replies under a specific post.
 - **Say nothing when you have nothing to add.** If a message doesn't need you,
-  don't reply. In a DM you may return the literal string `NO_REPLY` (and *only*
-  that string) to stay silent — never append `NO_REPLY` to real content, it will
-  be posted verbatim.
+  don't reply. In a DM return `NO_REPLY` as the entire reply to stay silent.
+  Do not begin substantive content with a bare `NO_REPLY`: direct API/MCP paths
+  suppress the whole reply, while gateway paths currently strip the token and
+  post the remainder. Backtick `NO_REPLY` when you need to mention it.
 - **In a 1:1 DM** you're talking to one peer — reply to every message, talk
   directly, and surface any shareable result to a team pod when you're done.
 

From 7bfccd66e9da882e15515d361fa1f8b132ce584c Mon Sep 17 00:00:00 2001
From: Lily Shen <115414357+lilyshen0722@users.noreply.github.com>
Date: Wed, 26 Aug 2026 00:59:03 -0700
Subject: [PATCH 06/10] fix(agents): retain leading sentinel audit context

---
 .../agentMessageService.chatNoise.test.js     | 15 +++++++++++++++
 backend/services/agentMessageService.ts       | 19 ++++++++++++++++---
 2 files changed, 31 insertions(+), 3 deletions(-)

diff --git a/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js b/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js
index 0d6c71c9d..14dfe28f9 100644
--- a/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js
+++ b/backend/__tests__/unit/services/agentMessageService.chatNoise.test.js
@@ -53,6 +53,10 @@ describe('sanitizeAgentContent — NO_REPLY suppression and sanitization', () =>
       .toBe('');
     expect(AgentMessageService.sanitizeAgentContent('\n\t NO_REPLY.\nHere is the real answer.'))
       .toBe('');
+    // Legacy gateways can concatenate duplicate silence blocks. A run is still
+    // a leading sentinel, not an ordinary word that should strip-and-post.
+    expect(AgentMessageService.sanitizeAgentContent('NO_REPLYNO_REPLY\nHere is the real answer.'))
+      .toBe('');
   });
 
   it('shares the silent-turn decision with non-posting consumers', () => {
@@ -233,6 +237,17 @@ describe('AgentMessageService.sanitizeAgentContent — strip observability', ()
     expect(line).toContain('pod=pod123');
     expect(line).toContain(input);
   });
+
+  it('retains a substantial audit excerpt for a suppressed authored turn', () => {
+    // The legacy 120-character cap suits reproducible runtime diagnostics. A
+    // direct API/MCP/CLI post suppressed here has no persisted turn body, so
+    // retain materially more of it for the audit record.
+    const body = 'x'.repeat(2048);
+    const input = `NO_REPLY\n${body}`;
+    expect(AgentMessageService.sanitizeAgentContent(input, OBSERVE)).toBe('');
+    const [line] = leadingSuppressionWarnings();
+    expect(line).toContain(body);
+  });
 });
 
 describe('AgentMessageService.postMessage — silent agent-DM conclusion hook', () => {
diff --git a/backend/services/agentMessageService.ts b/backend/services/agentMessageService.ts
index dc0d482ac..afe2e25f9 100644
--- a/backend/services/agentMessageService.ts
+++ b/backend/services/agentMessageService.ts
@@ -53,6 +53,15 @@ const RECENT_ATTACH_WINDOW_MS = 5 * 60 * 1000;
 // pattern it feeds, which matches at the start of a statement.
 const ATTACH_CLAIM_SCAN_LIMIT = 2000;
 
+// A leading bare NO_REPLY suppresses an authored reply before it is stored.
+// Native runtime turns retain their untruncated text in AgentRun before this
+// service is called, but direct API/MCP/CLI posts without an AgentRun have no
+// equivalent durable turn record. Keep a useful bounded audit excerpt for that
+// latter population rather than reusing the 120-character cap for reproducible
+// runtime-failure diagnostics. OpenClaw currently strips the sentinel upstream,
+// so its posts do not reach this branch.
+const LEADING_NO_REPLY_LOG_EXCERPT_LIMIT = 4096;
+
 // Runtime artifacts are exact values, not a language heuristic. Short,
 // all-caps agent replies regularly carry valid protocol or markup identifiers;
 // add an entry here only after observing it as a wrapper artifact in production.
@@ -1763,10 +1772,14 @@ class AgentMessageService {
 
     const trimmed = raw.trim();
     const sentinel = 'NO_REPLY';
-    const nextCharacter = trimmed[sentinel.length];
+    let sentinelEnd = 0;
+    while (trimmed.startsWith(sentinel, sentinelEnd)) {
+      sentinelEnd += sentinel.length;
+    }
+    const nextCharacter = trimmed[sentinelEnd];
     const nextIsWordCharacter = nextCharacter !== undefined
       && /[A-Za-z0-9_]/.test(nextCharacter);
-    return trimmed.startsWith(sentinel) && !nextIsWordCharacter;
+    return sentinelEnd > 0 && !nextIsWordCharacter;
   }
 
   /**
@@ -1877,7 +1890,7 @@ class AgentMessageService {
     if (AgentMessageService.hasLeadingBareNoReply(trimmed)) {
       if (observe) {
         console.warn(
-          `[agent-msg] suppressed substantive reply with leading bare sentinel from agent=${observe.agentName} instance=${observe.instanceId} pod=${observe.podId}: ${trimmed.slice(0, 120)}`,
+          `[agent-msg] suppressed substantive reply with leading bare sentinel from agent=${observe.agentName} instance=${observe.instanceId} pod=${observe.podId}: ${trimmed.slice(0, LEADING_NO_REPLY_LOG_EXCERPT_LIMIT)}`,
         );
       }
       return '';

From 966192b09cde51949a763d4e0544b3641bb16d49 Mon Sep 17 00:00:00 2001
From: Lily Shen <115414357+lilyshen0722@users.noreply.github.com>
Date: Wed, 26 Aug 2026 01:26:09 -0700
Subject: [PATCH 07/10] test(memory): cover leading sentinel history scan

---
 ...temExchangeTriggers.leadingNoReply.test.ts | 82 +++++++++++++++++++
 1 file changed, 82 insertions(+)
 create mode 100644 backend/__tests__/unit/services/systemExchangeTriggers.leadingNoReply.test.ts

diff --git a/backend/__tests__/unit/services/systemExchangeTriggers.leadingNoReply.test.ts b/backend/__tests__/unit/services/systemExchangeTriggers.leadingNoReply.test.ts
new file mode 100644
index 000000000..7322aa3c0
--- /dev/null
+++ b/backend/__tests__/unit/services/systemExchangeTriggers.leadingNoReply.test.ts
@@ -0,0 +1,82 @@
+// @ts-nocheck
+// TASK-067: the conclusion-memory scan must use the same leading-bare
+// NO_REPLY semantics as postMessage. This test keeps that cross-module
+// contract executable without standing up Postgres.
+
+const mockPool = { query: jest.fn() };
+
+jest.mock('../../../config/db-pg', () => ({ pool: mockPool }));
+jest.mock('../../../models/Pod', () => ({ findById: jest.fn() }));
+jest.mock('../../../models/User', () => ({
+  find: jest.fn(),
+  findOne: jest.fn(),
+}));
+jest.mock('../../../services/agentMemoryService', () => ({
+  appendSystemExchange: jest.fn().mockResolvedValue({ revision: 1 }),
+  truncateTakeaway: (value) => String(value).slice(0, 280),
+}));
+
+const Pod = require('../../../models/Pod');
+const User = require('../../../models/User');
+const { appendSystemExchange } = require('../../../services/agentMemoryService');
+const { recordAgentDmConclusion } = require('../../../services/systemExchangeTriggers');
+
+const POD_ID = '507f1f77bcf86cd799439011';
+const NOVA_USER_ID = '507f191e810c19729de860ea';
+const PIXEL_USER_ID = '507f191e810c19729de860eb';
+
+function queryResult(value) {
+  return {
+    select: jest.fn().mockReturnThis(),
+    lean: jest.fn().mockResolvedValue(value),
+  };
+}
+
+describe('recordAgentDmConclusion — leading-bare NO_REPLY history', () => {
+  beforeEach(() => {
+    jest.clearAllMocks();
+    Pod.findById
+      .mockReturnValueOnce(queryResult({ type: 'agent-dm' }))
+      .mockReturnValueOnce(queryResult({
+        type: 'agent-dm',
+        name: 'Nova and Pixel',
+        members: [NOVA_USER_ID, PIXEL_USER_ID],
+      }));
+    User.find.mockReturnValue(queryResult([
+      { username: 'nova', botMetadata: { agentName: 'openclaw', instanceId: 'nova' } },
+      { username: 'pixel', botMetadata: { agentName: 'openclaw', instanceId: 'pixel' } },
+    ]));
+    User.findOne.mockReturnValue(queryResult({ _id: NOVA_USER_ID }));
+  });
+
+  it('skips a stored leading-bare reply before recording the sender’s prior substantive takeaway', async () => {
+    mockPool.query.mockResolvedValue({
+      rows: [
+        { content: 'NO_REPLY\nThis suppressed body must not become the takeaway.' },
+        { content: 'The auth patch is ready for the final gate.' },
+      ],
+    });
+
+    await recordAgentDmConclusion({
+      podId: POD_ID,
+      senderAgentName: 'openclaw',
+      senderInstanceId: 'nova',
+      ts: new Date('2026-08-26T08:30:00Z'),
+    });
+
+    expect(mockPool.query).toHaveBeenCalledWith(
+      expect.stringContaining('WHERE pod_id = $1 AND user_id = $2'),
+      [POD_ID, NOVA_USER_ID],
+    );
+    expect(appendSystemExchange).toHaveBeenCalledWith(expect.objectContaining({
+      agentName: 'openclaw',
+      instanceId: 'nova',
+      takeaway: 'The auth patch is ready for the final gate.',
+    }));
+    expect(appendSystemExchange).toHaveBeenCalledWith(expect.objectContaining({
+      agentName: 'openclaw',
+      instanceId: 'pixel',
+      takeaway: '@nova: The auth patch is ready for the final gate.',
+    }));
+  });
+});

From d64a219eb545f1a3c63b0dd38e9cd1372bb9449a Mon Sep 17 00:00:00 2001
From: Lily Shen <115414357+lilyshen0722@users.noreply.github.com>
Date: Wed, 26 Aug 2026 01:31:37 -0700
Subject: [PATCH 08/10] refactor(memory): remove duplicate sentinel fallback

---
 backend/services/systemExchangeTriggers.ts | 36 ++++------------------
 1 file changed, 6 insertions(+), 30 deletions(-)

diff --git a/backend/services/systemExchangeTriggers.ts b/backend/services/systemExchangeTriggers.ts
index a1935ca64..38d242985 100644
--- a/backend/services/systemExchangeTriggers.ts
+++ b/backend/services/systemExchangeTriggers.ts
@@ -140,39 +140,15 @@ async function findPreviousNonSilentMessage(podId: string, senderUserId: string)
     // eslint-disable-next-line global-require, @typescript-eslint/no-require-imports
     const AMS = require('./agentMessageService') as {
       sanitizeAgentContent?: (s: unknown) => string;
-      isSilentNoReply?: (s: unknown) => boolean;
-      AgentMessageService?: {
-        sanitizeAgentContent?: (s: unknown) => string;
-        isSilentNoReply?: (s: unknown) => boolean;
-      };
-      default?: {
-        sanitizeAgentContent?: (s: unknown) => string;
-        isSilentNoReply?: (s: unknown) => boolean;
-      };
     };
-    const sanitize = (
-      AMS.sanitizeAgentContent
-      ?? AMS.AgentMessageService?.sanitizeAgentContent
-      ?? AMS.default?.sanitizeAgentContent
-    );
-    const isSilentNoReply = (
-      AMS.isSilentNoReply
-      ?? AMS.AgentMessageService?.isSilentNoReply
-      ?? AMS.default?.isSilentNoReply
-    );
+    const sanitize = AMS.sanitizeAgentContent;
+    // The CommonJS module exports the service class directly. If an isolated
+    // caller replaces it without this predicate, do not invent a third
+    // sentinel matcher and accidentally record a false takeaway.
+    if (typeof sanitize !== 'function') return null;
     for (const m of result.rows) {
       const raw = typeof m?.content === 'string' ? (m.content as string) : String(m?.content ?? '');
-      const trimmed = raw.trim();
-      const cleaned = typeof sanitize === 'function'
-        ? sanitize(raw)
-        : (typeof isSilentNoReply === 'function'
-          ? (isSilentNoReply(raw) ? '' : trimmed)
-          // The lazy import can be absent in isolated tests. Keep that
-          // backstop aligned with the live total-match + leading-bare rule.
-          : (/^(?:NO_REPLY\s*)+$/.test(trimmed)
-            || /^NO_REPLY(?:$|[^A-Za-z0-9_])/.test(trimmed)
-            ? ''
-            : trimmed));
+      const cleaned = sanitize(raw);
       if (cleaned) return cleaned;
     }
     return null;

From 1c0a63e5b597c6b106406aaaf7b0596d2f11282d Mon Sep 17 00:00:00 2001
From: Lily Shen <115414357+lilyshen0722@users.noreply.github.com>
Date: Wed, 26 Aug 2026 01:33:50 -0700
Subject: [PATCH 09/10] docs(agents): scope suppressed-turn audit retention

---
 backend/services/agentMessageService.ts | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/backend/services/agentMessageService.ts b/backend/services/agentMessageService.ts
index afe2e25f9..eeae7d363 100644
--- a/backend/services/agentMessageService.ts
+++ b/backend/services/agentMessageService.ts
@@ -55,11 +55,11 @@ const ATTACH_CLAIM_SCAN_LIMIT = 2000;
 
 // A leading bare NO_REPLY suppresses an authored reply before it is stored.
 // Native runtime turns retain their untruncated text in AgentRun before this
-// service is called, but direct API/MCP/CLI posts without an AgentRun have no
-// equivalent durable turn record. Keep a useful bounded audit excerpt for that
-// latter population rather than reusing the 120-character cap for reproducible
-// runtime-failure diagnostics. OpenClaw currently strips the sentinel upstream,
-// so its posts do not reach this branch.
+// service is called. MCP and CLI posts have no equivalent durable turn record,
+// so this audit line is their only copy of a suppressed turn. Keep a useful
+// bounded excerpt for that population rather than reusing the 120-character cap
+// for reproducible runtime-failure diagnostics. OpenClaw currently strips the
+// sentinel upstream, so its posts do not reach this branch.
 const LEADING_NO_REPLY_LOG_EXCERPT_LIMIT = 4096;
 
 // Runtime artifacts are exact values, not a language heuristic. Short,

From 9bb6d8b86938734cf42a440036682691d07d0c4c Mon Sep 17 00:00:00 2001
From: Lily Shen <115414357+lilyshen0722@users.noreply.github.com>
Date: Wed, 26 Aug 2026 01:34:40 -0700
Subject: [PATCH 10/10] docs(agents): clarify suppressed-turn audit scope

---
 backend/services/agentMessageService.ts | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)

diff --git a/backend/services/agentMessageService.ts b/backend/services/agentMessageService.ts
index eeae7d363..1b3714948 100644
--- a/backend/services/agentMessageService.ts
+++ b/backend/services/agentMessageService.ts
@@ -55,11 +55,13 @@ const ATTACH_CLAIM_SCAN_LIMIT = 2000;
 
 // A leading bare NO_REPLY suppresses an authored reply before it is stored.
 // Native runtime turns retain their untruncated text in AgentRun before this
-// service is called. MCP and CLI posts have no equivalent durable turn record,
-// so this audit line is their only copy of a suppressed turn. Keep a useful
-// bounded excerpt for that population rather than reusing the 120-character cap
-// for reproducible runtime-failure diagnostics. OpenClaw currently strips the
-// sentinel upstream, so its posts do not reach this branch.
+// service is called. Other HTTP-posting clients (direct API, MCP, CLI wrappers,
+// and cloud-codex) have no equivalent durable turn record, so this audit line
+// is their only copy of a suppressed turn. Keep a useful bounded excerpt for
+// that population rather than reusing the 120-character cap for reproducible
+// runtime-failure diagnostics. OpenClaw currently strips the sentinel upstream,
+// so its posts do not reach this branch; if that gateway adopts this rule, it
+// joins the same no-record population.
 const LEADING_NO_REPLY_LOG_EXCERPT_LIMIT = 4096;
 
 // Runtime artifacts are exact values, not a language heuristic. Short,