From sci-brain
Writes structured survey reports (literature review, technology assessment, SOTA, pros/cons) from a populated knowledge base. Final stage of the survey pipeline.
How this skill is triggered — by the user, by Claude, or both
Slash command
/sci-brain:survey-writerThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Write a structured survey/assessment report for a technology, research field, or platform. The output is a self-contained document suitable for internal decision-making, investor communication, or team onboarding.
Write a structured survey/assessment report for a technology, research field, or platform. The output is a self-contained document suitable for internal decision-making, investor communication, or team onboarding.
Pipeline: This skill is the final stage of the survey → download-ref → survey-writer pipeline. Run /survey first to build a literature KB, then /download-ref to fetch PDFs and render full-text markdown, then this skill to produce the report. The skill can also run standalone if a project KB already exists.
Trigger phrases: "write up the survey", "write a review", "technology assessment", "what is it / pros and cons / SOTA", "evaluate this field", "business relevance of X", "write a report on X".
Follow skills/_shared/writing-workflow.md for context loading, citation handling, gap-filling research, output format, diagrams, and finish checks.
/survey first — the review needs a grounded reference base.CLAUDE.md for company description, product portfolio, customer map, or team structure. If found, include "Business Relevance"; if not found, ask whether to include it.docs/discussion/user-profile.md or memory.articles/YYYY-MM-DD-<topic>-review.{md,typ,tex} or a project-specific path if the user prefers.template.typ (in this skill's directory, with template.bib): copy it to the output path and fill it in. It already encodes the § Report structure below and ships helper functions — section_box, stage (flow/era strips), proscons (per-approach unpaired strengths/limitations), compare_table, problem_table.Draft each section, show the user, get feedback before proceeding to the next:
Organize the review by technical approach. The bulk of the report explains the approaches, and the state of the art and trade-offs live inside each approach rather than in separate global sections. Do not write a standalone global "Pros and Cons" or "State of the Art" section — those belong per-approach (see § 2).
Define the topic in 2–3 paragraphs for a reader who has never heard of it. Cover:
Include a diagram showing the key concept (architecture, data flow, or the problem framing). Use grid + rect/box for side-by-side layouts to avoid CeTZ overflow issues. Keep text inside fixed-width box() elements.
This is the core of the review. Identify the main approaches / method families in the field (typically 3–6) and give one subsection per approach. Optionally open with a short field-wide timeline or landscape (a CeTZ timeline, or a one-line-per-era list) to orient the reader before the per-approach detail.
For each approach, cover three things in order:
@citekey citation. Lead with the best result, not a chronology.End the section with an optional cross-approach comparison table — rows = approaches, columns = a few shared criteria (e.g. scalability, verifiability/cost, maturity, best-fit use case), cells cited where they make a claim. Use it as the at-a-glance summary when the field has several comparable approaches; skip it for a single-approach topic.
Ranked table with columns: #, Problem, Why it matters, Who could solve it, Urgency (Critical / High / Medium). 4–8 rows. Order by urgency descending. For each problem, cite the paper(s) that define the gap or the closest existing work.
Structure:
rect with colored border)Use diagrams to make abstract concepts concrete. Prefer simple layouts over complex CeTZ canvases.
Typst reports: Use CeTZ (@preview/cetz:0.4.0) for timeline and dependency diagrams. For side-by-side comparisons and role diagrams, prefer Typst's native grid + rect + box — these handle text wrapping and overflow better than CeTZ content() nodes. Refer to skills/_shared/typst-reference.md for CeTZ patterns.
Common diagram types for reviews:
grid(columns: 3) with rect boxes and a center connectorline axis, circle milestones, content labelstable (rows = approaches, columns = shared criteria), not CeTZrect nodes (string names, not content), line arrows, color-coded by urgencyLayout rules:
content() with a fixed-width box() to prevent overflowname: parameters, never content blocksgrid instead of CeTZ — it handles text reflowEvery claim in tables should have at least one @citekey citation.
npx claudepluginhub quantumbfs/sci-brain --plugin sci-brainGenerates 50+ page market research reports with professional LaTeX formatting, data-driven visuals, and multi-framework analysis (Porter Five Forces, PESTLE, SWOT, TAM/SAM/SOM, BCG Matrix).
Generates 50+ page market research reports with consulting-firm style (McKinsey, BCG, Gartner), including LaTeX formatting, strategic frameworks (Porter Five Forces, PESTLE, SWOT, TAM/SAM/SOM, BCG Matrix), and data gathering via research-lookup.
Produces multi-source research reports via a CAFleet-orchestrated team (Director, Manager, Scouts, Researchers) with folder-based output.