Search and read ~11M full-text papers, FDA filings and clinical trials as a read-only virtual filesystem, with line-pinned citations.
Picks the governing framework (ICH Q2(R2)/Q14, M10, USP, CLSI, ISO 17025) first, then plans, computes and documents analytical method validation, verification and transfer.
Convert coordinates across BED/GFF/VCF formats, normalise variants, and catch assembly mismatches before they silently corrupt an analysis.
Build and audit reproducible NeuroKit2 pipelines for ECG, EDA, RSP, EEG and other physiological time-series analysis.
Ports CPU-bound scientific Python to NVIDIA GPUs and proves the result is both correct and actually faster.
A safety-first guide and CLI toolkit for reading, writing, converting and de-identification-preflighting local DICOM data with pydicom 3.0.2.
Design, audit, and export truthful, accessible, publication-ready scientific figures with Matplotlib, Seaborn, or Plotly.
Guides Claude through choosing SHAP explainers, computing and validating attributions, and plotting them correctly.
Anthropic's official skill for researching, building, testing, and evaluating high-quality MCP servers in TypeScript or Python.
Analyzes every open issue and PR in parallel and writes permalink-backed triage reports to /tmp — without ever mutating anything on GitHub.
Scans, LSP-verifies and deletes unused TypeScript code, delegating per-file batches to parallel agents that commit atomically.
Runs five specialized sub-agents in parallel — goal verification, hands-on QA, code quality, security, and context mining — and only passes the review when all five pass.
A router skill that loads the right taste, brand, performance, and design-ops rulesets so frontend work ships at senior-designer quality.
Fans out parallel research swarms across your codebase, the web, official docs, and OSS repos, then delivers a fully cited, empirically verified report.
A router skill that loads the right design, performance, and design-system rulesets so UI work ships at senior-designer quality instead of AI slop.
Rigorously verifies web and terminal UIs with screenshot/pixel-diff evidence plus two parallel read-only reviewers, then issues one GOOD/NEEDS WORK verdict.
Runs 3 vulnerability hunters and 2 PoC engineers in parallel so only findings with a proven attack path get reported.
Detects an existing OpenClaw install and interactively migrates identity, channel credentials, scheduled tasks, and config into NanoClaw v2.
Enforces strict typing, modern toolchains, and TDD discipline whenever Claude writes Python, Rust, TypeScript, or Go.
Splits a project's full claude-mem timeline into ISO weeks and writes one serialized narrative chapter per week, each inheriting the previous week's carry-forward.