example-summarize-notes

Example Skill: Summarize Notes with NotebookLM

Skill Purpose

Help an AI agent take a collection of notes (Markdown, text, transcripts) in this repo and produce a concise, grounded summary that can be safely added back to README.md or AGENTS.md.

When to Use This Skill

• You have 2–20 small notes, docs, or transcripts that all relate to the same topic.
• You want a short summary (1–3 paragraphs or 3–7 bullets) capturing key ideas for developers.
• You want the summary to be grounded in the source material, not hallucinated.

Inputs

• note_paths: list of file paths (relative to repo root) to summarize.
• summary_style (optional): one of bullets, paragraph, or mixed.
• target_audience (optional): short description, e.g. backend engineers, ML platform team.

High-Level Workflow

1. Collect the contents of all note_paths.
2. Optionally use an external grounding tool like NotebookLM for deeper analysis.
3. Ask the model to propose a candidate summary.
4. Cross-check the summary against the source notes; remove unsupported claims.
5. Emit a final summary plus a short "How to integrate into this repo" suggestion.

Detailed Agent Instructions

Follow these steps whenever this skill is invoked:

1. Load Inputs

   • Read all files listed in note_paths.
   • If any file is missing or empty, report which ones and continue with the rest.

2. Normalize Content

   • Strip boilerplate (license headers, navigation, TOCs) where obvious.
   • Preserve code snippets and CLI commands exactly; do not rewrite them.

3. Grounded Analysis (Optional NotebookLM Step)

   • If NotebookLM is available:
     • Create or update a NotebookLM notebook with the note contents.
     • Ask NotebookLM for 3–5 key takeaways and 3 concrete examples or use cases.
     • Use those responses as additional context, but do not copy phrasing verbatim.
   • If NotebookLM is not available, skip this step and rely on local analysis.

4. Draft Summary

   • If summary_style is bullets, produce 3–7 bullets.
   • If summary_style is paragraph, produce 1–3 short paragraphs.
   • If summary_style is mixed or not provided, use 1 short paragraph plus 3–5 bullets.
   • Target audience:
     • If target_audience is provided, emphasize what matters for that audience.
     • Otherwise, assume "software engineers adopting AI tooling".