# Episodic And Multi Hop Memory Reasoning

> Use this skill when the user wants questions like 'last time this happened', 'after they went there, what did they do next', 'combine clues from different parts', or 'connect the dots across a very long video.' Trigger it when the answer cannot be read from one clip and instead depends on stitching together multiple past moments.

- Skill: `dingxingdi/episodic-and-multi-hop-memory-reasoning-2` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx skillmds@latest add dingxingdi/episodic-and-multi-hop-memory-reasoning-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/dingxingdi/episodic-and-multi-hop-memory-reasoning-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: dingxingdi (https://skillmd.com/u/dingxingdi)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/dingxingdi/episodic-and-multi-hop-memory-reasoning-2

---


# Skill: episodic-and-multi-hop-memory-reasoning
## 1. Capability Definition & Real Case
* **Professional Definition**: The capability to answer questions whose evidence is distributed across distant segments, days, or episodes, requiring retrieval of multiple memories and a reasoning chain that links them into one conclusion.
* **Dimension Hierarchy**: Long-Horizon Reasoning->Global Content Integration->episodic-and-multi-hop-memory-reasoning
### Real Case

**[Case 1]**
* **Initial Environment**: A week-long egocentric video log with hierarchical summaries and retrievable subclips. The relevant supermarket escalator event occurred days before the question timestamp.
* **Real Question**: Who was in front of me last time I stood on the escalator in the supermarket?
* **Real Trajectory**: Retrieve the week-level memory using 'escalator' and 'supermarket', drill down to the matching day, inspect the identified escalator subclip, and connect the retrieved high-level memory about companions with the frame-level visual evidence.
* **Real Answer**: Tasha.
* **Why this demonstrates the capability**: No single local clip is enough unless the agent first finds the right day and the right escalator event. The problem therefore tests long-horizon episodic retrieval plus multi-hop reasoning over summary-level and frame-level evidence.

---
## 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`