🌩️ The STORM method
O STORM is a research method that originated at Stanford. The core idea is simple and powerful: instead of asking a question from just one angle, you build several deliberate perspectives on the same topic, let them disagree, and only then synthesize. In this course, the method becomes a skill called by Claude Code storm-research.
🧭 The idea in one sentence
Five specialists research the same topic in parallel, a map shows where they contradict each other, and each claim is checked against the original source before becoming a conclusion.
- •More perspectives → fewer blind spots.
- •Explicit contradiction → you can see where the evidence conflicts.
- •Verification → no invented numbers or studies.
🟡 New here? — “perspective” / “lens”
A lens (or perspective) is just a role the agent takes on — "act like an economist," "act like a skeptic." The same topic, viewed through different roles, reveals different facts. STORM has five fixed lenses, which you’ll learn about in Module 1.2.
Why this matters to you
If you create content, make technology decisions, or advise clients, a single-angle report leaves you exposed to what it missed. STORM is an inexpensive way to “borrow” expertise you don’t have and uncover your own blind spots.
Five angles, not one
Cross-check the perspectives
Primary source
Final deliverable
🕳️ The blind spot of a single prompt
When you trigger a single prompt, it uses a single framing — one way of looking at the problem. Anything that framing doesn’t consider important simply doesn’t make it into the research. The model isn’t lying; it’s answering the narrow question you asked.
The image below compares the two paths: on the left, a single prompt follows a narrow path and leaves gaps of coverage; on the right, five lenses cover complementary angles.
The red circles on the left show what the single angle didn’t cover. On the right, each lens (cyan) fills in a different angle on the same topic.
✗ What the single prompt does
- ✗Assumes one framing and ignores the rest
- ✗Agrees with your premise instead of testing it
- ✗Doesn't show where the evidence contradicts itself
✓ What multiple lenses do
- ✓Each lens covers an angle the others miss
- ✓The skeptic deliberately attacks the premise
- ✓The contradiction becomes part of the report, not an error
⚖️ Native deep research vs. STORM
Claude has a feature for deep research that launches many generic agents (sometimes more than a hundred) to scour the web. It's powerful, but tends to deliver a "dump" of statistics with few confirmed sources—and may hit usage limits. STORM makes the opposite bet: a few deliberate angles, each verified.
📊 Volume isn’t the same as structure
- Deep research: many identical agents → lots of text, overlap, few confirmed sources.
- STORM: ~9 to 11 agents with distinct roles → complementary coverage + citation verification.
- Practical consequence: STORM is more predictable in cost and less prone to rate limits.
More agents ≠ better research
One hundred agents that think alike find the same things a hundred times. Five that think differently find five distinct things.
Predictable by design
You know it will always be the same five lenses—the behavior and cost won’t unexpectedly explode.
🧩 What is a skill by Claude Code
A skill is, at its core, a master prompt saved in a file. When you invoke the skill, Claude reads this entire file and follows the instructions in it. The storm-research is exactly that: a file SKILL.md that describes the 4-phase pipeline.
🟡 New here? — three terms
- Claude Code: the version of Claude that runs in your terminal/editor and can use tools (read files, search the web, run commands).
- SKILL.md: the text file (Markdown) with the skill instructions. It’s just text—you can open and read it.
- Invoke: “calling” the skill. Just ask in natural language (“run a storm research on X”)—no special command needed.
Why “skill = prompt” is good news
Since it's just text, the skill is self-contained and portable: depends only on Claude Code's native tools—no external scripts, APIs, or paid services. You drop the folder into .claude/skills/ and it works. And because you read the file, you can adapt it (you do that in Track 3).
--- name: storm-research description: Use when someone asks you to run Storm Research… argument-hint: "[topic to research]" ---
📄 What you get at the end
The final deliverable is an standalone HTML file — a briefing that opens in your browser. It has a fixed structure that you’ll master in Track 3; for now, get to know the main parts.
The established fact, followed by the contested interpretation.
By reliability, with a 1–10 rating and “supports/challenges” chips.
The link that appears only when you cross-check the lenses.
Each citation has a status: confirmed, corrected, or downgraded.
💡 Practical tip
You can download the report template (report-template.html) and the skill in homepage download section to see the structure before you even run it.
🎚️ When to use it—and when it's overkill
STORM costs more than a quick search (about 9–11 agents). Use it when the topic has more than one defensible side and claims worth checking. For "what is the capital of France?", that's overkill.
✓ Good for STORM
- ✓"Are voice agents worth investing in?"
- ✓Topics with hype, disputed numbers, and incentives
- ✓Decisions where being wrong is costly
✗ Overkill / use something else
- ✗Simple, undisputed factual query
- ✗"Convert 50 miles to km"
- ✗When you just want a quick answer
Self-recovery (optional): which of these is the best use case for STORM?
📌 Module summary
Next module:
1.2 — The five expert lenses