knowledge-update-tracking

# Skill: knowledge-update-tracking ## 1. Capability Definition & Real Case * **Professional Definition**: The ability to detect when previously stored user information has changed and to privilege the updated state over stale memory without forgetting the underlying entity. * **Dimension Hierarchy**: Conversational Memory->Persistent Personal Memory->knowledge-update-tracking ### Real Case **[Case 1]** * **Initial Environment**: The assistant has prior sessions mentioning a family trip to Hawaii and later sessions mentioning a more recent family trip to Paris. Both remain in the history, and the user now asks about the most recent destination. * **Real Question**: Where did I go on my most recent family trip? * **Real Trajectory**: 1. Retrieve all family-trip mentions across sessions. 2. Sort them by recency rather than by salience or repetition. 3. Mark earlier destinations as valid historical memory but not the current answer target. 4. Return the latest destination only. * **Real Answer**: Paris. * **Why this demonstrates the capability**: This tests whether the assistant can update memory state instead of clinging to an earlier fact. The model must preserve both memories while still answering from the latest valid state, which is a common failure mode in long-lived assistants. ## 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`