Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →fradser/storm-engine
Provides the shared STORM methodology, artifact layout, stage-gating contract, citation hygiene, and retrieval fallback.
Passed 9 of 18 test cases with the skill, and 5 of 18 without it.
and 4 more in the full report
6 fixed5 passed either way10 still failing1 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
During the outline phase of STORM, two outline markdown files are created. Which file holds the initial unrefined parametric draft, and which file holds the research-refined structure consumed by the writing phase?
| Checked | Without skill | With skill |
|---|---|---|
| The parametric outline is saved in outline-draft.md and the research-refined outline is saved in outline.md | Missed | Passed |
The instructions your model reads when the skill loads.
Internal knowledge for the STORM plugin. Encodes the methodology ported from Stanford STORM (NAACL'24 / EMNLP'24) into a Claude-native pipeline. Loaded automatically by every /storm:* skill.
The pipeline has two stages and four phases. Each phase is independently runnable and resumable — completed phases are loaded from the artifact directory, never re-run.
Stage 1 (Pre-writing) Stage 2 (Writing)
research ---------> outline ---------> write ---------> polish
(persona (draft + (per-section, (summary +
discovery, refine from cited, parallel) dedup)
simulated Q&A, research)
information table)
NEVER skip stage 1. An article generated from parametric memory alone is not a STORM article — it will hallucinate citations and miss the multi-perspective grounding that defines the method. If research/ is absent, run /storm:research first.
NEVER invent citations. Every inline [n] must map to an entry in sources.json collected during research. If a section has no supporting sources, write it without citations and flag the gap in the polish phase — do not fabricate a reference.
All phases read from and write to a single per-topic directory:
<output_dir>/<slug>/
research/
personas.json # discovered personas (name + perspective + rationale)
conversations.jsonl # one record per (persona, turn): question, queries, snippets, answer
sources.json # deduplicated Information objects: {id, title, url, description, snippets}
outline.md # refined outline (markdown headings)
outline-draft.md # pre-research draft (kept for reference)
article.md # per-section draft with inline [n] citations
article-polished.md # final polished article
run-config.json # snapshot of run parameters
Stage-gating contract: a phase is considered complete when its primary artifact exists and is non-empty. Before running a phase, check its output artifact:
test cases passed, vs 5 / 18 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.