knhuang-research-skills

Scaffold a new research project: creates method.md, agenda.md, CLAUDE.md, and AGENTS.md with standard Big 5 folder structure. CLAUDE.md/AGENTS.md embed the lab's truth-over-harmony, coding-discipline, and research-integrity guardrails. Removes the friction of starting from scratch every time.

Use when drafting or revising scientific prose (proposals, paper intros, grant narratives, significance sections) and the sentences feel choppy, list-like, or jump between ideas — or when a reviewer says the writing 'doesn't flow.' Applies the known-new chaining technique distilled from Kangning Huang's NASA NESSF proposal: each sentence ends on a new idea that becomes the next sentence's opening topic (A→B, B→C), and each paragraph's first sentence reaches back to the previous paragraph's last idea. Keywords: flow, transitions, coherence, choppy, doesn't flow, topic chaining, known-new contract, anadiplosis, paragraph hinge.

Analyze causal relationships by searching for and evaluating scientific evidence. Use when the user asks whether X causes Y, inquires about causal mechanisms, or wants to distinguish causation from correlation. Triggers include questions like "does X cause Y", "is there a causal relationship between", "what's the evidence that X leads to Y", or "causation vs correlation" discussions.

Transform dense academic papers into compelling narratives using Malcolm Gladwell's storytelling techniques. Use when users ask to explain, interpret, summarize, or make accessible a research paper, academic article, or scientific study—especially when they want engaging, readable prose rather than technical summaries. Triggers include requests like "explain this paper," "help me understand this research," "make this accessible," "write about this study," or "Gladwell-style explanation."

Transform academic papers into pop-science summaries using the 'ladder-building' science-writing approach developed by Kangning Huang (start from what the reader already knows, then build up step by step). Requires WebSearch to research authors, Read for PDFs. Use when user shares a paper PDF or asks for a paper summary.

Strategic field reconnaissance for a domain you know a little but not enough. Produces an 8-dimension landscape map ending in a gap table ranked by a 3-filter comparative-advantage test against the user's own research pillars. Domain coverage: complexity science, urban science, urban sustainability, industrial ecology, economic geography. Scouting mode only — no prose drafting, no PDF ingestion. Conductor pattern: delegates verification and deep-read work to other Lu Lab skills.

Generate a print-ready HTML reading guide with prioritized chapter checklist and goal-targeted extraction questions. Takes a table of contents (image, PDF, paste, or URL) plus reading context, produces a single self-contained HTML file with tiered chapters, sub-topic checkboxes, and 3 to 5 guided questions per chapter. Use when the user has a book or paper to read and wants a structured plan before diving in.

Adversarial figure QA: reads the rendered figure AND its generating script to catch legend swaps, color-meaning errors, axis failures, arithmetic, and communication failures before the figure reaches anyone. Inspired by the lab's figure roasting tradition.

Scan writing for empirical claims (statistics, dates, historical facts, government data) and fact-check each via web search. Inserts verified source URLs inline. Distinct from manage-refs (which verifies DOIs). Use when reviewing drafts, grant narratives, blog posts, or any writing with factual claims.

Lean memory capture before /clear. Scans conversation for reusable patterns, appends max 5 bullet points to project MEMORY.md. Prevents memory bloat.

Adversarial review: simulate Reviewer #2 to find methodological gaps, logical leaps, missing literature, and statistical concerns before submission. Use on paper drafts, grant narratives, or analysis reports.

DOI-verified citation pipeline: resolve DOIs via CrossRef, build refs.bib, validate citation pools. Use when introducing ANY source not already in the Bib folder, or when building reference lists for grants/papers.

Trace a paper's intellectual roots: find the 1-2 deepest knowledge trunks it recombines. Based on the Innovation Depth principle — innovation magnitude scales with recombination depth. Use standalone or as an enrichment layer for /read-ken or any bib-note pipeline.

Project warm-up briefing: reads agenda.md (and CLAUDE.md if present) to summarize recent progress, state where we left off, and flag blockers. Use when restarting work on any coding or writing project.

Brainstorm diagnostic and characteristic figures for an analysis. Designs a visual GUI — single-variable distributions, bivariate relationships, and refined multi-facet views — so an outsider can evaluate the analysis without reading code.