| name | multi-perspective-scan |
| description | Use this skill when the user wants a fast multi-angle read of a topic without the full STORM pipeline — phrases like "multi-perspective scan", "give me 5 perspectives on <topic>", "analyze <topic> from multiple angles", "what would different experts say about <topic>", "perspective scan", "5 expert views", "steelman the different takes on <topic>", or "what are the different viewpoints on <topic>". Fans out the five STORM persona agents (practitioner, skeptic, economist, historian, academic) in parallel and presents their views. This is STORM Phase 1 standalone — the reusable heart of the method.
|
| version | 1.0.0 |
| user-invocable | true |
Multi-Perspective Scan — The Five Lenses, Standalone
When you ask one question, you get the majority view — one framing, the surface. The
breakthrough behind Stanford's STORM is simple: ask the same topic from five independent
expert lenses and you catch the blind spots a single pass never sees. This skill runs just
that fan-out — fast, when you don't need the full contradiction map / synthesis / peer-review
pipeline (for that, use storm-research).
The five lenses, and what each uniquely catches:
| Lens | Sees what others miss |
|---|
| Practitioner | The theory–practice gap; what actually works in the field |
| Skeptic | Overclaims, hidden flaws, the buried failure cases |
| Economist | Who profits, who pays, the misaligned incentives |
| Historian | The pattern that repeated before; where we are in the cycle |
| Academic | What the evidence actually shows, including conflicting findings |
Workflow
Step 0 — Scope
- Get the topic. If vague or sprawling, ask one sharpening question. A tight topic yields
sharper perspectives.
- Derive a slug and set the workspace
.storm/{slug}/. Create it: mkdir -p .storm/{slug}.
- Tune the panel if a swap clearly helps (e.g. add a Clinician / Regulator / End User via a
Task call using the same persona output contract). Default to the standard five.
Step 1 — Fan Out (parallel)
Launch all five persona agents in a single message with five Task calls. Each gets the
topic, the workspace path, and its output file. Per-agent prompt:
"Topic: {topic}. Workspace: .storm/{slug}/. Write your analysis to
.storm/{slug}/{persona}.md per your output contract. Gather real evidence with
WebSearch/WebFetch and cite it."
Targets: practitioner.md, skeptic.md, economist.md, historian.md, academic.md.
Each agent writes its file and returns a one-line confirmation. If one fails, note it and
continue with the rest.
Step 2 — Present the Five Views
Read the five files and present a compact digest — the bottom line and 2–3 sharpest insights per
lens, plus where you already see them diverging:
Multi-Perspective Scan — {topic}
=================================
Workspace: .storm/{slug}/
🛠 Practitioner — {bottom line}
• {top insight} • {top insight}
🔍 Skeptic — {bottom line}
• {top flaw [FATAL/CAVEAT]} • {…}
💰 Economist — {bottom line}
• {top incentive} • {…}
📜 Historian — {bottom line}
• {top parallel} • {…}
🎓 Academic — {bottom line}
• {top finding [STRONG/MIXED/…]} • {…}
Early signal: {one line on the most interesting disagreement you already notice}
Full analyses: .storm/{slug}/{persona}.md
Step 3 — Offer the Next Step
Suggest the natural follow-ups:
- Map where they disagree → run
contradiction-map on this workspace
- Go all the way to a briefing → run
storm-research (adds synthesis + peer review)
- Drill into one lens → ask that persona a follow-up
Notes
- This is intentionally lighter than
storm-research: five views, no synthesis. Use it to scan
fast, then escalate only if the topic warrants the full pipeline.
- The workspace it produces is directly consumable by
contradiction-map and storm-research
(Phases 2–4), so a scan is never wasted work.
