💭 in your head · nothing to type
You have your pipeline design from the last unit — a sketch of how the steps connect.
Don't look at it as a diagram. Picture it as plumbing: every connection is a fitting the flow has to pass through. Where pipes meet tight, water runs clean. Where they're joined in a hurry — it drips.
Run each joint of your design in your head: which one honestly looks like a temporary clamp held together with tape? That's the fitting to reinforce before you turn the water on.
💭 in your head · nothing to type
Picture an editor at the cutting table: hundreds of clips in front of them. The goal isn't to make a perfect film — it's to assemble a first rough cut so they can see where the story falls apart.
No great edit starts from a first take. First comes the rough cut: everything in order, even if it's uneven. Then the director watches to see what grips and what sags.
Your pipeline today is the rough cut. Notice the one spot that sagged on the first pass — that's exactly where the second version gets its upgrade.
This is a working prompt you can use right now:
You are an analytical agent. Your job is to extract maximum value from content.
URL to analyze: [paste link]
Tasks:
1. Read the page fully
2. Pull out 5–7 actionable ideas (concrete, not generic phrases)
3. Identify 5–7 topics for further research
4. Find 2–3 best quotes or facts
Output format:
## Key ideas
[list with explanations]
## Topics to research
[list with short context]
## Worth remembering
[quotes or facts]
If you can't open the link, say so; don't summarize from guesswork.
In a chat (claude.ai) the model won't create a file: copy the result into your ideas file, e.g. my-experiments/ideas.md. An agent with file access (Claude Code) can write it itself — add "Save the result to my-experiments/ideas.md" to the prompt.
If the model can't open the link (this happens in a chat without a scraping tool), don't accept an answer based on guesswork. Copy the article text and paste it into the message instead of the URL.
You can give several URLs in a single request:
Analyze three articles on the topic [topic]:
1. [URL 1]
2. [URL 2]
3. [URL 3]
For each: 3 ideas + 1 best quote.
Then: find what they have in common and where they contradict.
If you couldn't open a link, name it and don't analyze it.