Thinking Notebook

This skill separates collecting from deciding. The agent collects and proposes, the person decides what stays. It is meant for conference days, meetings, streams, and learning phases where more input arrives than can be remembered.

The basis is the thinking notebook from the keynote "Mensch oder Maschine?" ("Human or Machine?") by Roger Basler de Roca at the AI Superevent Vienna 2026. The personalizable starter kit with prompt, automation, and Notion template is available at https://nachdenken.fragroger.ai/en

Output language is American English, unless the person wants something else.

When this skill applies

• A transcript, a recording, or dictated notes need to be processed.
• After a conference day or talk: "What do I take away from this?"
• The person wants to build their own thinking notebook as an artifact.
• Kept notes should turn into thoughts or a daily recap.

Two modes

Clarify at the start which mode is meant:

1. Process (default): input in, notes out, person filters, three thoughts, export. Works in any chat tool.
2. Build: The agent builds the notebook as an interactive artifact with persistent storage. Needs a tool with HTML output and storage.

Process mode

Step 1: Capture the entry

Per meeting or talk, one entry with:

• Name of the speaker or participants
• Title
• Time
• Source (transcript, own notes, event app, stream)

Accepted formats: pasted text, .txt, .md, .vtt, .srt. For .vtt and .srt, remove timestamps and numbers, keep speaker changes. If the title is missing, ask once.

Step 2: Pull out the notes

Each note is a single statement from the input that is understandable on its own. Volume is welcome, the person filters themselves. Each note gets exactly one type:

Symbol	Type	Trait
💡	Insight	Realization, thesis, new perspective
📊	Number	Concrete number, study, measurement with context
💬	Quote	Verbatim from the text, in quotation marks
✅	Task	Something the person could or should do
❓	Open question	Unresolved point, contradiction, needs more research
🤖	AI agent	An agent shown or described, with a level

Levels for AI agents:

Level	Meaning
1	reads
2	drafts
3	acts with approval
4	acts autonomously

If the level is not clear from the text, set [level unclear] instead of guessing.

Step 3: Keep or strike

Output the notes as a numbered list and let the person decide:

NOTES · [Title] · [Name] · [Time]
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Delivered: 14 · Kept: 0 · Struck: 0

 1 💡 [Note]
 2 📊 [Number with context]
 3 💬 "[verbatim quote]"
 4 🤖 [Agent], level 3 (acts with approval)
 5 ✅ [Task]
 6 ❓ [Question]
...

Reply like: keep 1, 3, 5 · strike 2 (reason: already know this)

For every strike, ask for the reason if none was given. The reasons are part of the learning process and are carried along. Update the count above after every decision.

Step 4: Think back

Only once at least three notes have been kept:

• Draw three notes at random ("Reshuffle") and phrase each as an open thought, as a question or a tension, not as a solution.
• Ask the person to rewrite each thought in their own words.
• On request, add three suggested thoughts, clearly marked as a proposal:

SUGGESTED THOUGHT (Proposal)
Thesis:        [one sentence]
Evidence:      [notes from at least two different entries, with number]
Consequence:   [one sentence]

Build suggested thoughts only from kept notes. If all notes come from a single entry, say so openly.

Step 5: Export

Output as Markdown for Notion (import into Notion via Settings, Import, Text and Markdown):

# The Day in Notes · [Weekday], [MM/DD/YYYY]
> As of: [Time] · [n] entries · [n] notes · [n] kept, [n] struck

## Think ahead
- What was on the agenda today: ...

## Think along
### [Time] · [Title] · [Name]
Summary: [one sentence]
- **💡 [kept note]**
- ~~📊 [struck note]~~ (reason: ...)

## Think back
### Three suggested thoughts (proposal)
1. Thesis · Evidence · Consequence
### My three thoughts
1.
2.
3.

Kept notes in bold, struck notes with strikethrough and reason. Do not rephrase note text in the export.

Build mode

Build an artifact with these specs:

• Look: white background like a notebook, fine lines, handwritten headings, calm and easy to read. Three pages side by side, stacked on mobile.
• Storage: all entries persist. Name the data areas fixed: news, eingaenge, merker, verarbeitet, gedanken. At the end output the link and these names so that later scheduled tasks can access them.
• Think ahead: today's appointments if a calendar is connected, otherwise a field "What's on today?". A news area that Claude fills later.
• Think along: input for text or file (.txt .md .vtt .srt) with name, title, time. On save, generate notes per Step 2. Per note the buttons "Keep" and "Strike", with a field for the reason when striking. On top, the count delivered, kept, struck.
• Think back: "Reshuffle" draws three thoughts from kept notes, each editable. Button "Notion export" downloads everything as Markdown per Step 5.

Optional as scheduled tasks once the notebook is running:

• Mornings: research current news on set topics, each with a real URL, source, and one sentence, and replace the news area.
• Evenings: per day, create or update a Notion page "The Day in Notes", only with today's entries, never overwrite pages of other days.

Every scheduled task starts with no memory. The link to the notebook therefore belongs in every prompt.

Rules

• Invent nothing. Every note must trace back to a spot in the input.
• Quotes only verbatim. If the wording is uncertain (e.g. poor transcript), carry it as an insight instead of a quote.
• Numbers only with the context in which they were stated.
• The agent collects and proposes. Keeping, striking, and rewriting the thoughts is done by the person.
• Do not drop notes in advance because they seem trivial. Filtering is the person's job.
• Do not process confidential customer, HR, or health data. Transcripts of others only with their consent.

Anti-patterns

• Delivering a summary instead of individual notes.
• Presorting notes as "important" yourself and preempting the decision.
• Handing over the three thoughts fully phrased without asking for a rewrite.
• Suggested thoughts with evidence not found in the kept notes.
• Simply leaving struck notes out of the export. The reasons are part of the value.
• Guessing agent levels.

Self-check for the person

Anyone who only clicks "Keep" is no longer processing, just collecting. Then take a break and summarize the talk in three sentences yourself before reopening the notes.

Quality check before delivery

• Every note has exactly one type and is understandable on its own
• Quotes are verbatim, numbers have context
• AI agents have a level or [level unclear]
• The count delivered, kept, struck is correct
• Suggested thoughts are marked as a proposal and backed with evidence
• Export contains kept and struck notes with reason

---

Last updated: October 9, 2026 · Version 1.0.0 · skills.fragroger.ai

License: CC BY-NC-ND 4.0 (https://creativecommons.org/licenses/by-nc-nd/4.0/) with additional permission (CC+): use within companies and for client work allowed, adaptation for personal use allowed. Not allowed: selling, bundling, publishing modified versions. Details: https://skills.fragroger.ai/en/license