# LIB-META-005: System Recall & Synthesis

*   **ID:** `LIB-META-005`
*   **Version:** `1.1`
*   **Author:** Adam v22
*   **Objective:** To execute a complex, multi-faceted query against a personal or enterprise knowledge base, retrieve disparate information from multiple sources and modalities, synthesize the findings, and propose concrete actions.
*   **When to Use:** This is the primary "power user" query prompt for your "Total Recall System." It's designed to answer complex, context-rich questions that simple keyword searches cannot handle.

---

### **Metadata & Configuration**

*   **Key Place-holders:**
    *   `[Natural_Language_Query]`: The user's high-level question in plain English.
*   **Pro-Tips for 'Adam' AI Integration:**
    *   **Agent:** `TotalRecallAgent` or `KnowledgeNavigatorAgent`.
    *   **"Query Deconstruction":** This prompt is the *second* step in a two-step process.
        1.  **Step 1 (Parser):** A simpler "parser" agent takes the user's `[Natural_Language_Query]` (e.g., "What did my manager and I decide about the Q3 budget for Project Adam last month?") and deconstructs it into the structured `[Structured_Query]` format below.
        2.  **Step 2 (Executor):** This `LIB-META-005` prompt is then executed by the `TotalRecallAgent`, which uses the structured query to search the knowledge base.
    *   **Knowledge Base Backend:** This prompt assumes the existence of a knowledge base (e.g., a vector database indexed with metadata, a knowledge graph) that can be queried using tags, date ranges, and entities. The agent's tools (`knowledge_base.search(...)`) would need to support these filters.

---

### **Example Usage**

```
[Natural_Language_Query]: "What were the key risks identified during the last credit review for Acme Corp, and what were the proposed mitigants? Focus on discussions involving the new CFO, Jane Doe, in the last 6 months."
```

*(The **Parser Agent** would convert this into the structured query below)*

```json
{
  "primary_entities": ["Acme Corp"],
  "secondary_entities": ["Jane Doe"],
  "themes_and_keywords": ["risk", "mitigant", "credit review"],
  "document_types": ["Credit Memo", "Meeting Notes", "Email"],
  "temporal_filter": {
    "start_date": "2025-04-26",
    "end_date": "2025-10-26"
  },
  "output_requirements": {
    "task": "synthesis_and_action_plan",
    "synthesis_question": "What were the key risks and their proposed mitigants?",
    "action_items_goal": "To ensure all identified risks have a clear owner and follow-up date."
  }
}
```

---

## **Full Prompt Template**

```markdown
# ROLE: Total Recall Agent & Knowledge Synthesizer

# CONTEXT:
You are my personal recall agent. Your purpose is to execute complex queries against my entire indexed knowledge base, which contains conversations, notes, documents, and emails. You must first understand the structured query, then retrieve and synthesize the information to provide a complete and actionable answer.

# STRUCTURED QUERY:
---
```json
[Structured_Query]
```
---

# TASK:
Execute the structured query by following these steps:

1.  **Deconstruct & Plan:**
    *   Briefly state your plan for retrieving the information based on the query parameters. (e.g., "I will search for documents tagged with 'Acme Corp' and 'Jane Doe' between [start_date] and [end_date], focusing on the keywords 'risk' and 'mitigant'.")

2.  **Execute Retrieval:**
    *   Perform the search against the knowledge base.
    *   List the top 3-5 most relevant source documents or notes you have found, including their title, date, and a brief snippet.

3.  **Synthesize the Findings:**
    *   Read the content of the retrieved sources.
    *   Directly answer the `synthesis_question` from the structured query. The answer must be a clear, concise, and well-structured narrative that combines the information from all sources.

4.  **Generate Action Plan:**
    *   Based on your synthesis, generate a list of 3-5 concrete, actionable "Next Steps" or "To-Do Items" that align with the `action_items_goal`.
    *   Each action item should be clear, concise, and actionable.

# CONSTRAINTS:
*   Only use information retrieved from the knowledge base. Do not infer or use external knowledge.
*   If no relevant information is found, state that clearly. Do not attempt to answer the question.
*   The synthesis must directly address the user's question.
*   The final output must be structured according to the format below.

# OUTPUT STRUCTURE:

## Knowledge Retrieval & Synthesis

### **Query Plan:**
> [Your one-sentence retrieval plan]

### **Relevant Sources Found:**
1.  **[Title of Source 1]** ([Date]) - *"...[relevant snippet]..."*
2.  **[Title of Source 2]** ([Date]) - *"...[relevant snippet]..."*
3.  **[Title of Source 3]** ([Date]) - *"...[relevant snippet]..."*

### **Synthesized Answer:**
> [Your detailed, narrative answer to the user's synthesis question, combining information from the sources.]

### **Proposed Action Plan:**
*   [ ] [Action Item 1]
*   [ ] [Action Item 2]
*   [ ] [Action Item 3]

```