agentsclimarketplace

Storm research

Skill josherau/claude-operating-core/skills/storm-research

Use when researching any topic, question, market, company, technology, or decision — before writing reports or proposals, making business or investment decisions, preparing for negotiations, interviews, or presentations, or learning a new field. This is the default research method for all research tasks. Skip only for single-fact lookups, or when the user explicitly asks for a quick answer or names a different method.From its SKILL.md

Install
npx -y skills add josherau/claude-operating-core --skill storm-research

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

4.0 KB, 810 tokens by cl100k_base, as published. Nobody here has run it

STORM Research

Overview

Multi-perspective research method adapted from Stanford's STORM (Shao et al., NAACL 2024, github.com/stanford-oval/storm). Core principle: one query returns the majority view; five grounded perspectives plus a contradiction map returns understanding. The STORM paper reports ~25% better organization and ~10% broader coverage from multi-perspective question asking versus its single-pass baseline.

The Iron Rule: every perspective must be grounded in real retrieval. Persona-prompting without sources is five flavors of one model's blind spots hallucinating in harmony. Each perspective's claims come from actual searches (WebSearch, WebFetch, or whatever search MCPs you have wired up) with sources noted. No search → no claim.

The Four Phases

Run all four, in order. Do not stop after Phase 1 or 3.

Phase 1 — Multi-Perspective Scan

Pick 5 perspectives suited to the topic. Default set:

PerspectiveAsks
PractitionerWhat actually happens day-to-day? What breaks in reality?
SkepticWhat's overhyped or wrong? What could fail?
EconomistWho profits? What incentives shape the claims?
HistorianHas this pattern happened before? How did it end?
AcademicWhat does the actual evidence/data say?

Swap perspectives when the topic demands (e.g. regulator, customer, competitor, clinician). For each: generate that persona's 2-3 sharpest questions, then search to answer them. Note the source for each finding.

Phase 2 — Contradiction Map

Find where the perspectives fight and why. Rules of evidence:

  • All 5 agree → probably true.
  • Perspectives conflict → the real understanding lives here; name the disagreement and the reason (incentive, timeframe, evidence quality).
  • Nobody addressed it → you found a gap; say so explicitly.

Phase 3 — Synthesis

One briefing: what's established (with sources), the named contradictions, reliability ranking of major claims, and a specific recommended action — not a summary that hedges everything.

Phase 4 — Peer Review

Adversarially grade your own output: strongest claim, weakest claim, likely source bias, missing perspective, and a confidence score per key conclusion. The original STORM pipeline has no self-critique step, so this phase is our addition — and it is not optional. (And per the review-panel rule: Phase 4 is still self-critique — the finished deliverable gets an independent reviewer too.)

Scaling

ModeWhenHow
QuickCasual question, minutes matter3 perspectives, ~5 searches, all inline
Standard (default)Most research requests5 perspectives, 8-15 searches, inline or 2-3 parallel subagents
Deep"Thorough/comprehensive/audit"One subagent per perspective in parallel, then synthesize; consider a dedicated deep-research harness if you run one

Output Shape

Deliver: Briefing (synthesis + recommended action) → Contradiction mapPerspective findings with sourcesPeer review with confidence scores. Briefing first — the reader gets the answer before the apparatus.

Common Mistakes

  • Answering from training data with persona flavoring. The perspectives structure the questions; retrieval answers them.
  • Perspectives that converge. If the skeptic sounds like the academic, you picked lazy personas — force genuine disagreement in framing.
  • Skipping Phase 2. The contradiction map is the step that separates this from a book report.
  • Skipping Phase 4 to save time. 60 seconds of self-critique is the method's error bar.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most research analysis skills give in 810 tokens

Counted across 1,213 of the 2,113 authors here whose files we hold, read 2026-09-06

  • Cite sources for every important claimin 47 of 1213, across 38 files
  • Separate facts from inferences and recommendationsin 21 of 1213, across 12 files
  • Write findings to a markdown filein 19 of 1213
  • Label every insight with a confidence levelin 18 of 1213, across 8 files
  • Read product marketing context before asking questionsin 18 of 1213, across 8 files
  • Rank themes by frequency and intensityin 16 of 1213, across 6 files
  • Establish research mode before proceedingin 16 of 1213, across 6 files
  • Segment survey responses by customer tier or tenurein 16 of 1213, across 6 files
  • Categorize support tickets before analyzingin 16 of 1213, across 6 files
  • Weight research sources from the last twelve monthsin 16 of 1213, across 6 files
  • Use at least five data points per segmentin 15 of 1213, across 5 files
  • Extract verbatim quotes for all research findingsin 15 of 1213, across 5 files

Said here and by no other author read

  • Select five distinct research perspectives
  • Generate sharp questions for each persona
  • Search to answer all generated questions
  • Note sources for every finding
  • Identify conflicts between perspectives
  • Explain reasons for identified disagreements

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.