A collection of skills for LLM agents (Claude Code and others).
An interactive REPL-style assistant for scientific writing. Supports journal articles, abstracts, conference papers, grant proposals, and research blog posts. Enforces one-sentence-per-line formatting for clean git diffs, proposes changes before committing, and never hallucinates facts.
Documents a data directory rigorously, producing README.md, PROVENANCE.md, MANIFEST.tsv, DOWNLOAD.sh, and TEST_DOWNLOAD.sh. Verifies provenance against sources actually fetched, records tiered checksums with their authority, and tests the download script. Gaps are marked UNVERIFIED and reported rather than guessed.
Instantiates a bioinformatics analysis project repository: directory scaffold, AGENTS.md standards with CLAUDE.md symlinked to it, a dataset registry with provenance READMEs, the payload/ vs gate/ access protocol with dummy-data workflow, pinned environments, an experiment template ending in a readable report, and a literature-ingestion pipeline that normalizes papers to structured JSON (JATS/XML → HTML → JSON + figures → PDF fallback) with a schema and validator. Seeds literature, notes, and data during instantiation, then stops before any analysis.
Designs and refactors readable, comprehensive pytest unit tests for ProteoPy's Python and proteomics code. Organizes each function's contract in a single test class and covers happy paths, boundary conditions, validation, invariants, and AnnData-specific behavior with focused fixtures, parametrization, and precise assertions.