scicomp-research-skills

Improve academic paper writing quality for ML/CV/NLP-style papers with clear section structure, paragraph flow, and reviewer-facing presentation. Use when drafting or revising Abstract, Introduction, Related Work, Method, Experiments, or Conclusion; polishing figures/tables; checking claim-support alignment; or performing self-review before submission.

Author or revise any project document whose primary audience is a HUMAN -- README.md, PLAN.md, per-paper survey notes (notes/survey_*.md), bibliography collection logs (references/_collection_log.md), reviewer-response drafts, per-section research notes (notes/section_*.md), per-component implementation plans (notes/impl_*.md). Apply the human/agent audience split (these docs are NOT downstream renderings of AGENTS.md), the two-tier readability structure (orient-quickly + look-up-later), proper sectioning + cross-references, narrative prose over telegraphic fragments, and tables/trees/code-blocks where they aid scanning. Use whenever the agent will produce text that a human collaborator (the user or a co-author or a reviewer) is expected to read for review or reference.

Use this skill whenever a session will touch a scientific-computing codebase (numerical methods, PDE solvers, inverse problems, OED, UQ, scientific ML, or any code that produces numbers) in Python, Julia, C++, or other languages. Make sure to load this EVEN IF THE USER DOES NOT MENTION IT. Codifies eleven disciplines for AI-assisted scientific software development: numerical correctness (MMS, convergence-rate tests, conservation invariants, "paper tests" guard); testing strategies for numerical code; API design for researchers (NumPy / JAX / dolfinx / petsc4py idioms); performance + scaling; reproducibility infrastructure (lockfiles, Zenodo); CI/CD; project lifecycle; code-paper coupling (commit-pinning, submission tags); numerical-launch + debug protocols; and the Bridgeford et al. 2025 ten rules for AI-assisted coding in science. Cites Scientific Python Development Guide (BSD-3), pyOpenSci, Wilson et al. 2017, JOSS 2025 criteria, The Turing Way. Composes with `literature-survey` + `research-paper-writing`.

Use this skill at the start of any non-trivial session in this ecosystem -- the single biggest lever for reducing token / quota / context-window consumption and for keeping the agent productive across sessions. Make sure to load this whenever the work involves more than a few file reads, any PDF handling, multi-file editing, web fetching, or continuation of prior-session work, EVEN IF THE USER DOES NOT MENTION IT. Codifies five disciplines: tool selection (Read/Grep/Glob/Edit/Write over Bash equivalents); PDF lifecycle (one-shot pdftotext + survey-note-first lookup); persistent memory via the project's indices (PLAN.md / collection log / notes index); context-window budgeting; web-fetch caching. A research-flavoured operationalisation of the file-as-memory + just-in-time retrieval patterns from Anthropic's context-engineering guidance and the Manus / planning-with-files / claude-mem prior art (see "Adjacent prior art + lineage" inside the skill for citations).

Build and maintain a verified literature-survey corpus for a research paper -- BibTeX entries, published-manuscript PDFs, text extractions for AI consumption, per-paper survey notes (~30-50 lines each), and a collection log tracking verification status and corrections. Use when starting a new paper that needs a substantial literature review, when adding new references to an existing paper, or when migrating an unverified bib file to a verified one.