long-horizon-information-recall-and-personalization
Use this skill when the user expects the assistant to remember something said much earlier and to use that memory in a helpful, personalized way. Trigger it for requests like 'remind me what you recommended before', 'what did I say about my setup', or 'suggest something that fits what you already know about me'. Everyday examples include: 'what restaurant did you recommend in Rome?', 'what accessories fit my camera kit?', 'what was my commute length again?', and 'use my earlier preferences in this new recommendation.'
来源信息
- 仓库
- Dingxingdi/paper_fast_search_backup
- 最近来源活动
- 2026年4月8日 15:14
- 检测到的 SKILL.md 语言
- 英语
- 星标
- 0
- 分支
- 0
安装方式
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
检查来源文件
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
文件资源管理器
4 个文件正在显示 SKILL.md
SKILL.md
来源说明 · 只读预览- name
- long-horizon-information-recall-and-personalization
- description
- Use this skill when the user expects the assistant to remember something said much earlier and to use that memory in a helpful, personalized way. Trigger it for requests like 'remind me what you recommended before', 'what did I say about my setup', or 'suggest something that fits what you already know about me'. Everyday examples include: 'what restaurant did you recommend in Rome?', 'what accessories fit my camera kit?', 'what was my commute length again?', and 'use my earlier preferences in this new recommendation.'
# 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`
在 GitHub 查看