Problem

When autoRecall is enabled, the plugin automatically injects relevant memories into the user prompt via <relevant-memories> tags before the agent processes the message. However, users have no visibility into what memories are being injected.

This creates a trust and accuracy risk:

• Old/inaccurate memories may silently influence the agent's behavior
• Users cannot verify or correct injected content
• The <relevant-memories> tag is stripped from persisted messages by the before_message_write hook, making it invisible in conversation history

Proposed Solution

Add a config option recall.showInjected (boolean, default false):

{
  "recall": {
    "showInjected": true
  }
}

When enabled, the before_message_write hook should skip stripping <relevant-memories> tags, preserving them in the conversation history so users can see what was injected.

Current Workaround

I patched dist/index.mjs to add this check:

const showInjected = api.pluginConfig?.recall?.showInjected ?? false;
api.on("before_message_write", (event) => {
    if (showInjected) return; // preserve <relevant-memories>
    // ... existing strip logic
});

This works but is lost on plugin upgrades.

Use Case

For long-term collaborative agents, memory accuracy degrades over time. Users need transparency into what the agent "remembers" to catch and correct stale or incorrect memories before they influence responses.