Agent-Skills.md

Agent Skills: Extract insights from a note

Extract actionable lessons, key insights, emergent todos, and best practices from a note (e.g. a video transcript or article in an Obsidian vault, for example). Use when asked to "extract insights / lessons / takeaways" from a note or document. Outputs bullet lists in chat, or writes them to a markdown file if asked.

UncategorizedID: fredoliveira/obsidian-skills/extract-insights

Repository

fredoliveira/obsidian-skills
fredoliveiraLicense: MIT
241

Install this agent skill to your local

pnpm dlx add-skill https://github.com/fredoliveira/agents/tree/HEAD/skills/extract-insights

Skill Files

Browse the full folder contents for extract-insights.

Download Skill

Loading file tree…

skills/extract-insights/SKILL.md

Skill Metadata

Name
extract-insights
Description
Extract actionable lessons, key insights, emergent todos, and best practices from a note (e.g. a video transcript or article in an Obsidian vault, for example). Use when asked to "extract insights / lessons / takeaways" from a note or document. Outputs bullet lists in chat, or writes them to a markdown file if asked.

Extract insights from a note

Purpose

Read one note (transcript, article, clipping, meeting notes — anything) and pull out what's worth keeping, as bullet lists under four headings:

  • Actionable lessons — things to do differently, stated as guidance.
  • Key insights — the non-obvious ideas worth remembering.
  • Todos — concrete next actions that emerge for the reader/user.
  • Best practices — repeatable rules or standards worth adopting.

Instructions

  1. Read the note in full. Take the path the user gave (default to the file they @-referenced or the active note). Page through long files — don't extract from a partial read.

  2. Extract. For each of the four headings, write a bullet list. Rules:

    • Omit a heading entirely if the note genuinely has nothing for it. Don't pad.
    • One idea per bullet, declarative and self-contained (readable without the source open).
    • No fixed count necessary. Capture what's real, skip sponsor reads, filler, and biography.
    • Quote a memorable line only when it earns its place.

  3. Output — composable, decided by how you were asked:

    • Default (no output file mentioned): print the bullet lists directly in the chat. Nothing is written to disk.

    • If the user asks for a markdown file (e.g. "save it", "create a note", or gives a path): write the lists to a .md file.

      • Title / filename: Insights from <source title>.md — keep it consistent so these notes are easy to spot.
      • Location: if the user gave a path, use it. Otherwise default to an Inbox/ folder at the vault root if one exists (so the user can triage it later), falling back to the source note's own folder.
      • Body: link back to the source note in the first line, then the lists.
      • Include light frontmatter:
      ---
source: "[[<source note title>]]"
created: <today's date, YYYY-MM-DD>
tags:
  - insights
---

# Insights from <source title>

Extracted from [[<source note title>]].

## Actionable lessons
- ...

## Key insights
- ...

      (Drop empty sections here too.) Then tell the user the path you wrote.

Notes

  • Keep it minimal. The model already knows how to extract — this skill just fixes the four buckets, the bar for inclusion, and the chat-vs-file output choice.
  • Adapt to the vault: if the user's notes use different tags or folders, match them.