ie-engine

Inspiration Engine: Orchestration and Attention Management

This skill is the unified entry point and orchestration center of the Inspiration Engine. It is responsible for connecting three layered skills, running the complete inspiration generation pipeline, and managing when and how to present insights to the user.

Applicable Scenarios

Passive Trigger (explicit user request)

The user says "help me see what inspiration there is in my past notes"
The user says "connect the dots" or "help me string my ideas together"
The user is doing creative work and wants to get inspiration from historical notes
The user wants to comprehensively review the evolution of their thinking in a certain field

Active Trigger (the Agent judges that the user may benefit)

When you observe the following signals, you should actively consider running the Inspiration Engine and remind the user:

The user is deeply discussing a topic, and that topic may have related content in historical notes
The user has proposed an idea or hypothesis, and historical notes may contain supporting or contradictory evidence
The user repeatedly mentions similar concepts in different conversations, suggesting that this is a long-term focus
The user is making a decision or planning, and historical experience may provide valuable reference

During active triggering, do not directly run the complete pipeline. Instead, first give the user a lightweight one-sentence prompt, and let the user decide whether to go deeper. See "Active Trigger Protocol" below for details.

Inapplicable Scenarios

If the user only needs a single capability, guide them to the corresponding layered skill:

Only need to retrieve notes --> ie-retrieve-memory
Only need to analyze note relationships --> ie-connect-dots
Only need to generate insights --> ie-generate-insight
Create or edit notes --> directly use the wpsnote tool
General Q&A --> answer directly

Pipeline Flow

The complete inspiration generation pipeline contains three stages, executed in order:

阶段一：记忆检索（ie-retrieve-memory）
  输入：用户的当前上下文/主题/关键词
  输出：相关历史笔记列表
  ↓
阶段二：想法连接（ie-connect-dots）
  输入：阶段一的检索结果
  输出：聚类、连接假设、长期模式
  ↓
阶段三：洞见生成（ie-generate-insight）
  输入：阶段二的推理结果
  输出：洞见、探索建议、演化路径

For the detailed pipeline definition and parameter configuration, please read references/pipeline.md。

Orchestration Execution Steps

Step 1: Understand the Context

Before starting the pipeline, first clarify the user's intent and current context:

Determine the topic or field the user cares about
Determine the user's expectations (comprehensive scan vs. focus on a certain direction)
If the information is insufficient, ask the user for supplementary information

Step 2: Run Memory Retrieval

Following the guidance of ie-retrieve-memory, retrieve relevant historical knowledge from the user's notes.

Parameter Control:

Comprehensive scan mode: retrieve 10-15 notes, using multiple retrieval strategies
Focused mode: retrieve 5-8 notes, concentrated on a specific topic

Step 3: Run Idea Connection

Pass the retrieval results to ie-connect-dots, and perform reasoning analysis according to its guidance.

Parameter Control:

Enable all three capabilities by default (clustering, connection, pattern recognition)
If the retrieval results are fewer than 3 items, skip clustering and only perform connection hypotheses.

Step 4: Run insight generation

Pass the reasoning results to ie-generate-insight, and generate inspiration content according to its guidance.

Parameter control:

By default, generate 2-3 insights + 2-3 exploration suggestions.
If there are clear clues of temporal evolution, additionally generate an idea evolution path.

Step 5: Attention management and presentation

Before presenting the results to the user, make an attention management judgment.

Proactive trigger protocol

When the Agent proactively determines that the user may benefit from historical notes, follow the protocol below:

Step 1: Quick pre-check

Do not immediately run the full pipeline. First use ie-retrieve-memory to perform a lightweight retrieval (2-3 notes) to quickly determine whether there is relevant content.

Step 2: Lightweight prompt

If the pre-check finds relevant notes, remind the user in one sentence, in the following format:

[灵感提示] 你之前在「[笔记标题]」等 [N] 条笔记中讨论过相关的想法，要我帮你深入看看有什么启发吗？

Key principles:

Use only one sentence; do not expand into analysis.
Mention the specific note titles so the user can judge whether they are relevant.
End with a question, leaving the decision to the user.
Do not interrupt the workflow the user is currently working on.

Step 3: Wait for the user's response

The user says affirmative replies such as "Okay" or "Let me take a look" --> run the full pipeline.
The user ignores it or says "Not for now" --> do not ask again and continue the current work.
The user says "Remind me later" --> remember this topic and give a lightweight prompt again at an appropriate time.

The principle of restraint for proactive triggering

In the same conversation, proactive prompts should appear at most 2 times to avoid disturbing the user.
If the user has already declined a proactive prompt once, reduce the sensitivity of subsequent triggers.
When the user is clearly in a focused execution state (continuously issuing specific instructions), do not insert prompts.

Attention management

The inspiration engine should not bombard the user with information. Before presenting insights, consider the following factors:

Presentation judgment conditions

Only when the following conditions are all met should the complete insight results be presented:

Novelty: The insight contains connections the user may not be aware of (not obvious ones).
Confidence: At least one connection hypothesis has "medium" or "high" confidence.
Actionability: At least one exploration suggestion can be acted on immediately.

Handling of different situations

All three conditions are met: Present the complete results (insights + exploration suggestions + evolution path).
Only some conditions are met: Present a streamlined version, showing only the parts that meet the conditions.
No valuable findings: Tell the user frankly and suggest adjusting the retrieval direction or expanding the scope.

Presentation format

When presenting to the user, use the following structure:

灵感报告
===

检索范围：[使用的关键词/主题，涉及多少条笔记]

---

[洞见1]
[洞见2]（如果有）

---

下一步探索建议
[建议列表]

---

[想法演化路径（如果有）]

---

说明：以上洞见基于对你过去笔记的分析，是可能的启发方向，不代表确定性结论。

Architecture overview

ie-engine（本技能 - 编排与注意力管理）
  |
  +-- ie-retrieve-memory（记忆层 - 检索历史知识）
  |
  +-- ie-connect-dots（推理层 - 分析想法关系）
  |
  +-- ie-generate-insight（洞见层 - 生成灵感内容）

Each layered skill can be used independently. The value of this skill lies in:

Providing a one-click complete pipeline experience
Managing data transfer between layers
Performing attention management and controlling output quality