thought

Thought — Capture Surprising Insights

Capture a model answer that surprised you during an exploration conversation. The skill synthesizes the answer with its context, saves it as a granular entry, and updates the TL;DR synthesis file.

This skill is non-interrupting. After the thought is saved, resume the conversation exactly where it was — do not change topic or add commentary about the save.

Directory: ~/thoughts/ — created on first save; set the THOUGHTS_DIR environment variable to override.

When to Activate

• User triggers /thought during a conversation
• The user just read something in your last answer that surprised them — something they were NOT thinking before

Steps

Step 1: Synthesize the Last Answer

Look at your most recent substantive answer in the conversation (the one just before the user triggered /thought).

Produce a synthesis with two parts:

1. Context (3-5 sentences max):

   • What the exploration was about (the thread/topic)
   • The specific question or angle that led to the surprising answer

2. Thought (3-8 sentences max):

   • The synthesized insight from the model's answer
   • Focus on what was surprising or non-obvious — not a generic summary
   • Write in a way that will make sense when re-read weeks later without the full conversation

Step 2: Present for Confirmation

Show the synthesis to the user in this format:

**Context:** [context text]

**Thought:** [thought text]

**Suggested subject:** [1-2 word subject tag, or "none"]

Then ask: "Save this? You can: accept, reframe (tell me what to change), or cancel. You can also change or skip the subject."

Wait for the user's response. If they ask to reframe, adjust and present again. If they cancel, stop — resume conversation.

Step 3: Save the Entry

On confirmation:

1. Determine the filename:
   • Format: YYYY-MM-DD-subject-slug.md (e.g., 2026-04-06-emergent-patterns.md)
   • If no subject: YYYY-MM-DD-untitled.md
   • If filename already exists, append -2, -3, etc.