Qinyan Academic Skills
An installable library of AI agent skills covering literature discovery, scientific writing, grants, bioinformatics and statistics, giving mainstream coding agents end-to-end research capabilities.
Evidence shows upstream attribution is reasonably handled: README credits K-Dense-AI/claude-scientific-skills, luwill/research-skills, Yuan1z0825/nature-skills, and states per-skill provenance lives in each SKILL.md (2). However, install.sh and any SKILL.md are absent from the reviewed files, so least privilege, side effects, sensitive-data handling, user confirmation, and rollback cannot be verified and score thin (1). The curl | bash pattern lacks checksums or signatures — deducted on external effects. No red-line risk (malware, exfiltration) is evident in the provided files.
Dependencies (Bash, Git, curl) are stated with Windows guidance (2), but the README claims 187 skills while the object description says 182 — an internal inconsistency — and no failure-message behavior is documented (1).
Audience (academic researchers), six supported agents, and global vs project scopes are clearly delineated (2); trigger precision and per-skill capability boundaries rest on SKILL.md files not provided (1).
Information architecture is strong (category table, consistent naming, five localized READMEs) (2); install notes and examples are thorough (2); MIT license with per-skill caveat is clear (2). No changelog/versioning, no known-limitations section, no maintenance commitment — deducted (1).
Selective install, search, status, and update lifecycle lower cost of use, and plain-text skills ease review (2); actual output quality is not assessable statically (not penalized here).
Upstream project names give partial traceability, but headline figures (187 skills, 6 agents, category counts) cannot be corroborated within the provided files, and claims/inference separation is weak (1).
- The object description states 182 skills while the README claims 187; verify with --list-skills before installing.
- curl | bash executes a remote script without checksums or signatures; review install.sh before running in security-sensitive environments.
- Publisher is unverified in the FollowAgents registry — treat identity as unknown, not suspicious.
- Individual skills may carry additional or differing license terms; check each SKILL.md before redistribution or commercial use.
- No SKILL.md or install.sh source was available for this static review, so trust and behavior conclusions are low-confidence.
What does this agent do, and when should you use it?
Qinyan Academic Skills (GitHub: LeonChaoX/qinyan-academic-skills) is a multilingual collection of AI agent skills, with 187 installable skills across 18 domains per the README. It ships a first-party five-skill Nature-style suite (writing, polishing, review, figures, statistics) alongside curated scientific skills derived from open-source projects. An install.sh script targets six agents — Claude Code, Cursor, Codex, Gemini CLI, OpenClaw, and OpenCode — with global or per-project installation and full/category/single-skill granularity plus lifecycle management (search, status, updates). Every skill is plain text with a SKILL.md entry point, making it reviewable and customizable. The project is MIT-licensed; the five first-party Qinyan skills rely on the Qinyan Academic OpenAPI for paper search and related services. Note a discrepancy: the repository description says 182 skills while the README says 187 — verify with --list-skills after install.
install.sh writes skill files into a target agent's skills directory (e.g. ~/.claude/skills/ for Claude Code). Flags like --skill, --category, --tool, and --project control scope; --list/--list-skills/--search/--status/--check-update/--update handle discovery and maintenance. Each skill is a SKILL.md text instruction set: the Nature suite produces evidence-led narratives, meaning-preserving polished prose, prioritized review findings with stable issue identifiers, reproducible multi-panel figures, and design-aware statistical plans; qinyan-paper-search queries Google Scholar, PubMed, arXiv, and Wanfang via the Qinyan Academic OpenAPI; other skills encode usage of tools such as Scanpy, Biopython, RDKit, and statsmodels.
- A PhD student uses qinyan-paper-search and literature-review to systematically search PubMed and arXiv and map evidence.
- A researcher runs qinyan-nature-polishing and qinyan-nature-review for language polish and traceable pre-submission review.
- A lab team drafts grant proposals with nsfc-proposal and research-grants skills.
- A computational biology group uses scanpy, scvi-tools, and pydeseq2 for single-cell and differential expression analysis.
- A medicinal chemistry team applies rdkit, diffdock, and molecular-dynamics for molecular processing and docking.
- A clinical researcher uses clinicaltrials-database and clinical-decision-support for trial and decision-support workflows.
What are this agent's strengths and limitations?
- One installer targets six agents (Claude Code, Cursor, Codex, Gemini CLI, OpenClaw, OpenCode), avoiding vendor lock-in
- Selective global/project installation by skill or category, with search, status, and update lifecycle commands
- The five-skill Nature-style suite covers the full submission chain from argument and review to figures and statistics, with traceable issue identifiers
- Skills are version-controllable plain text, easy to review, customize, and share
- First-party skills like qinyan-paper-search depend on the Qinyan Academic OpenAPI; quota, key, and pricing are undocumented
- No native Windows support — the installer requires WSL or Git Bash
- Skill count is inconsistent between the repo description (182) and README (187), so documentation needs verification
- "Nature-style" denotes editorial goals only; no affiliation with Nature Portfolio and no acceptance guarantee
- Curated third-party skills may carry per-skill license terms in their SKILL.md that must be checked before redistribution
How do you install or deploy this agent?
Requires Bash, Git, and curl (WSL or Git Bash on Windows). Full install: curl -fsSL https://raw.githubusercontent.com/LeonChaoX/qinyan-academic-skills/main/install.sh | bash (defaults to the global Claude Code skills directory). Alternatively git clone https://github.com/LeonChaoX/qinyan-academic-skills.git and run bash install.sh --help.
How do you use this agent?
Scope as needed: curl -fsSL "$INSTALLER" | bash -s -- --category 01 for a category; --skill scanpy for one skill; --project --skill scientific-writing to install into the current project; --tool codex for another agent (short flags -c/-s/-t supported). Maintain with bash install.sh --list-skills, --search "protein", --status, --check-update, and --update. Start with one skill's SKILL.md and combine skills only when the workflow needs it.