Set up
Skills are installed by placing their folder where your agent loads skills.
The download comes from github.com/AI4S-YB/fan-skill itself, not from our server, so what you get is exactly what its author published.
Install instructions come from the README
README
synced Oct 26, 2026fan-skill
AI-powered plant bioinformatics and breeding analysis engine. From biological question to publication-ready results. Three-layer knowledge architecture (personal → general → Agent fallback) with B+C dual-mode decision system.
🇨🇳 中文文档
Quick Start
git clone git@github.com:AI4S-YB/fan-skill.git
cd fan-skill
# Install for your AI coding agent:
bash install-claude.sh # Claude Code
bash install-codex.sh # Codex CLI
bash install-gemini.sh # Gemini CLI
bash install-opencode.sh # OpenCode
bash install-all.sh # Auto-detect and install for all
# Or use the universal installer:
# npx skills add AI4S-YB/fan-skill
Then describe your biological question:
"I have 300 rice accessions with GBS data and 3 years of yield. Find genes controlling grain weight."
Fan-skill will match your intent to the knowledge base, check your experimental design, recommend an analysis path, and execute it.
📖 Quick Start Guide · Analysis Capabilities · Contributing
Architecture
User says "find genes controlling grain weight"
│
▼
┌─────────────────────────────────┐
│ SKILL.md │ ← Unified entry point
│ Intent → Three-Layer Match → Execute
└──────────────┬──────────────────┘
│
┌──────────┼──────────┐
▼ ▼ ▼
┌────────┐ ┌───────┐ ┌───────┐
│ User │ │ gwas │ │rnaseq │ ← user-knowledge/ (your experience)
│ entry │ │ rules │ │ rules │ knowledge-base/ (31 curated)
└───┬────┘ └───┬───┘ └───┬───┘
│ │ │
└──────────┼─────────┘
▼
┌─────────────────────────────────┐
│ engine/ │ ← Shared execution engine
│ rule_engine · validate · log │
└─────────────────────────────────┘
│
┌──────────┼──────────┐
▼ ▼ ▼
分析报告 决策日志 图表
│
(未匹配时) Agent 兜底推理 → 沉淀到 user-knowledge/
Validated on apple RNA-seq and rice candidate gene analysis. Each decision is auditable.
Three-Layer Knowledge
| Layer | Path | Priority | Content |
|---|---|---|---|
| User | user-knowledge/ | 🔺 Highest | Your personal analysis experience. Starts empty, grows with use. |
| General | knowledge-base/ | 🔸 Default | 31 curated entries covering major plant bioinfo workflows. |
| Agent Fallback | (model reasoning) | 🔻 Last resort | When neither layer matches, the Agent reasons and can precipitate results into your user layer. |
B+C Dual-Mode
| Layer | Role | Form |
|---|---|---|
| C-layer (Rule) | Deterministic — same data → same method | rules.yaml |
| B-layer (Expert) | Flexible — expert reasoning for edge cases | notebook.md |
| User control | `decision_mode: rule | expert |
| User entries | Minimum meta.yaml + notebook.md; optional rules.yaml | user-knowledge/ |
Capabilities
| Goal | Fan-skill can |
|---|---|
| Find trait genes | GWAS · QTL mapping · eQTL · candidate gene association |
| Understand function | RNA-seq DE · time-series · GRN · multi-omics |
| Predict breeding value | Genomic selection · hybrid prediction · enviromics |
| Process raw data | Variant calling · genotype imputation · phenotype analysis |
| Epigenomics | ChIP-seq · ATAC-seq · DNA methylation |
| Microbiome | 16S/ITS amplicon · metagenomics |
| Metabolomics & Proteomics | LC-MS/GC-MS · DIA/DDA · PPI |
| Evolution | Comparative genomics · pan-genome |
| Edit & breed | CRISPR design · marker development |
| Visualize | Publication-quality ggplot2 figures |
Full catalog: Analysis Capabilities
Design
Fan-skill encodes judgment knowledge — how an expert decides what analysis to run and which method to use. Not fixed pipelines.
- Pipeline problem: fixed steps fail on diverse data
- Atomic skill problem: free composition creates unstable invocation paths
- Fan-skill solution: Three-layer knowledge architecture. Your personal experience (
user-knowledge/) takes priority over the 31 curated entries (knowledge-base/). When neither matches, Agent reasoning serves as fallback and can be saved as new personal entries — the system grows with you. - B+C dual-mode: C-layer encodes "what method" as structured rules; B-layer encodes "why" as expert reasoning. Both layers coexist in every entry.
Every decision is auditable — engine/log_decision.sh records why each choice was made.
Contribute
To the general knowledge base: New analysis = 4 files (not 14). See Contributing Guide.
cp templates/rules-template.yaml knowledge-base/<name>/rules.yaml
cp templates/notebook-template.md knowledge-base/<name>/notebook.md
# + consult-guide.md + analysis-primer.md
engine/validate_entry.sh knowledge-base/<name>/
To your personal knowledge base: Just use fan-skill. When Agent fallback solves a novel analysis, choose to save it — an entry is auto-generated in user-knowledge/drafts/. Review and confirm to move it to user-knowledge/confirmed/.
License
MIT