Long Horizon Information Recall And Personalization

Skill: long-horizon-information-recall-and-personalization

1. Capability Definition & Real Case

• Professional Definition: The ability to recall user- or assistant-provided details from distant prior interactions and use them accurately to answer factual questions or personalize later responses.
• Dimension Hierarchy: Conversational Memory->Persistent Personal Memory->long-horizon-information-recall-and-personalization

Real Case

[Case 1]

• Initial Environment: The assistant has accumulated dozens of prior user-assistant sessions spanning work, travel, food, and photography. Relevant memory items are buried in older conversations rather than repeated in the current session.
• Real Question: Can you remind me of the romantic restaurant in Rome you recommended for dinner?
• Real Trajectory: 1. Search prior sessions for recommendation history tied to Rome and romantic dinner context. 2. Distinguish assistant-provided information from user-provided preferences. 3. Retrieve the specific restaurant name rather than paraphrasing the recommendation broadly. 4. Answer directly and, if appropriate, lightly personalize the response using the earlier preference context.
• Real Answer: Roscioli.
• Why this demonstrates the capability: This tests whether the assistant can recover a distant assistant-side memory item and use it in a user-facing way. The challenge is both recall accuracy and the ability to turn retrieved history into a personalized, context-aware reply without hallucinating adjacent details.

Pipeline Execution Instructions

To synthesize data for this capability, you must strictly follow a 3-phase pipeline. Do not hallucinate steps. Read the corresponding reference file for each phase sequentially:

1. Phase 1: Environment Exploration Read the exploration guidelines to discover raw knowledge seeds: references/EXPLORATION.md

2. Phase 2: Trajectory Selection Once Phase 1 is complete, read the selection criteria to evaluate the trajectory: references/SELECTION.md

3. Phase 3: Data Synthesis Once a trajectory passes Phase 2, read the synthesis instructions to generate the final data: references/SYNTHESIS.md