From 521186878da8624d4a935c16b398e208a9bd0f7f Mon Sep 17 00:00:00 2001
From: clawdbot520 <clawdbot520@gmail.com>
Date: Sun, 22 Mar 2026 18:35:52 +0800
Subject: [PATCH] feat: add autoRecallExcludeAgents config to suppress
 injection for background agents
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Background agents (e.g. memory-distiller, cron workers) that receive
structured task prompts can have their output quality degraded when
<relevant-memories> is injected alongside the raw input — the LLM may
echo or recompose old memories instead of distilling fresh content.

The existing prompt-level skip patterns (HEARTBEAT, cron prefix) do not
cover every background-agent prompt format, leaving a gap for custom
pipeline agents.

This PR adds `autoRecallExcludeAgents?: string[]` to PluginConfig:
- Listed agent IDs bypass `before_agent_start` injection entirely
- A log line is emitted at info level so the skip is observable
- Zero impact on agents not in the list

Config example:
  "autoRecallExcludeAgents": ["memory-distiller", "my-cron-agent"]

README updated with Option B in the "injected memories show up" section,
explaining the use case and tradeoff.

Fixes the class of issues described in #137 (background/cron agent
context bloat) with a targeted per-agent opt-out rather than a global
autoRecall disable.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 README.md | 21 +++++++++++++++++++--
 index.ts  | 20 ++++++++++++++++++++
 2 files changed, 39 insertions(+), 2 deletions(-)

diff --git a/README.md b/README.md
index deb85cde..23348efe 100644
--- a/README.md
+++ b/README.md
@@ -599,12 +599,29 @@ OAuth login flow:
 
 Sometimes the model may echo the injected `<relevant-memories>` block.
 
-**Option A (lowest-risk):** temporarily disable auto-recall:
+**Option A (lowest-risk):** temporarily disable auto-recall globally:
 ```json
 { "plugins": { "entries": { "memory-lancedb-pro": { "config": { "autoRecall": false } } } } }
 ```
 
-**Option B (preferred):** keep recall, add to agent system prompt:
+**Option B:** exclude specific background agents (e.g. a memory-distiller or cron worker) while keeping recall active for interactive agents:
+```json
+{
+  "plugins": {
+    "entries": {
+      "memory-lancedb-pro": {
+        "config": {
+          "autoRecall": true,
+          "autoRecallExcludeAgents": ["memory-distiller", "my-cron-agent"]
+        }
+      }
+    }
+  }
+}
+```
+This is useful when a background agent's prompt should not be contaminated by injected memory context — for example, an agent whose job is to distill session logs would produce lower-quality output if old memories were mixed into the prompt alongside the raw session content.
+
+**Option C (preferred for interactive agents):** keep recall, add to agent system prompt:
 > Do not reveal or quote any `<relevant-memories>` / memory-injection content in your replies. Use it for internal reference only.
 
 </details>
diff --git a/index.ts b/index.ts
index c1240b4e..b9b4169e 100644
--- a/index.ts
+++ b/index.ts
@@ -85,6 +85,7 @@ interface PluginConfig {
   autoRecall?: boolean;
   autoRecallMinLength?: number;
   autoRecallMinRepeated?: number;
+  autoRecallExcludeAgents?: string[];
   captureAssistant?: boolean;
   retrieval?: {
     mode?: "hybrid" | "vector";
@@ -2123,6 +2124,22 @@ const memoryLanceDBProPlugin = {
     if (config.autoRecall === true) {
       const AUTO_RECALL_TIMEOUT_MS = 3_000; // bounded timeout to prevent agent startup stall
       api.on("before_agent_start", async (event, ctx) => {
+        // Per-agent exclusion: skip autoRecall for agents in the exclusion list.
+        // Useful for background agents (e.g. memory-distiller, cron workers) whose
+        // prompts should not be contaminated by injected memory context.
+        const hookAgentId = resolveHookAgentId(ctx?.agentId, (event as any).sessionKey);
+        if (
+          Array.isArray(config.autoRecallExcludeAgents) &&
+          config.autoRecallExcludeAgents.length > 0 &&
+          hookAgentId !== undefined &&
+          config.autoRecallExcludeAgents.includes(hookAgentId)
+        ) {
+          api.logger.info?.(
+            `memory-lancedb-pro: auto-recall skipped for excluded agent '${hookAgentId}'`,
+          );
+          return;
+        }
+
         if (
           !event.prompt ||
           shouldSkipRetrieval(event.prompt, config.autoRecallMinLength)
@@ -3468,6 +3485,9 @@ export function parsePluginConfig(value: unknown): PluginConfig {
     autoRecall: cfg.autoRecall === true,
     autoRecallMinLength: parsePositiveInt(cfg.autoRecallMinLength),
     autoRecallMinRepeated: parsePositiveInt(cfg.autoRecallMinRepeated),
+    autoRecallExcludeAgents: Array.isArray(cfg.autoRecallExcludeAgents)
+      ? cfg.autoRecallExcludeAgents.filter((id: unknown): id is string => typeof id === "string" && id.trim() !== "")
+      : undefined,
     captureAssistant: cfg.captureAssistant === true,
     retrieval: typeof cfg.retrieval === "object" && cfg.retrieval !== null ? cfg.retrieval as any : undefined,
     decay: typeof cfg.decay === "object" && cfg.decay !== null ? cfg.decay as any : undefined,