Bio Agent Skills Hub

Bio Agent Skills Hub

A router/index skill over 1,676 deduplicated biomedical AI agent skills, aggregated and deduplicated from 20 open-source repositories into 15 categories. Each skill is a self-contained SKILL.md folder compatible with Claude-based agent frameworks.

• Repository: https://github.com/BioTender-max/awesome-bio-agent-skills
• Machine-readable index (authoritative): https://raw.githubusercontent.com/BioTender-max/awesome-bio-agent-skills/main/bioskill_index_v3.csv (1,676 rows)

This skill does not duplicate the 1,676 skills. It tells the agent how to (1) search the index, (2) locate the single best-matching skill, (3) fetch that skill's SKILL.md on demand, and (4) follow it. Skill bodies are fetched from GitHub on demand; a git clone fallback (incl. a China mirror) is provided for offline / bulk / private use.

How to use this skill (router workflow)

When a user asks for any bioinformatics task, workflow, or analysis:

Step 1 — Search the index to find candidate skills

The index has columns: skill_name, folder_name, source_repo, category, description, file_count, archive_path. archive_path = <source_repo>/<folder_name> and is the path under skills/.

Search skill_name + description + category + source_repo together (descriptions contain rich "Use when..." text), and present ranked candidates with their category and archive_path so the right one can be chosen:

import pandas as pd

# Authoritative v3 index. Prefer a local copy if this skill bundles one; else read from GitHub raw.
INDEX_URL = "https://raw.githubusercontent.com/BioTender-max/awesome-bio-agent-skills/main/bioskill_index_v3.csv"
df = pd.read_csv(INDEX_URL)   # 1,676 rows

def search_skills(query, category=None, source=None, top=15):
    """Keyword search over name+description+category+source. Returns candidate skills."""
    q = query.lower()
    hay = (df["skill_name"].fillna("") + " | " +
           df["description"].fillna("") + " | " +
           df["category"].fillna("") + " | " +
           df["source_repo"].fillna("")).str.lower()
    hits = df[hay.str.contains(q, regex=False)].copy()
    if category:
        hits = hits[hits["category"] == category]
    if source:
        hits = hits[hits["source_repo"] == source]
    # rank: name match first, then description match
    hits["_name_hit"] = hits["skill_name"].str.lower().str.contains(q, regex=False)
    hits = hits.sort_values(["_name_hit", "file_count"], ascending=[False, False])
    return hits[["skill_name", "category", "source_repo", "archive_path", "description"]].head(top)

print(search_skills("variant calling").to_string(index=False))

If nothing matches, broaden the query (try a synonym), or filter by category (see the table below) and browse that category's README section.

Step 2 — Fetch the chosen skill's SKILL.md (content on demand)

Once you pick a candidate's archive_path (e.g. bioskills/clair3-variants):

import urllib.request

archive_path = "bioskills/clair3-variants"   # from the search result
url = f"https://raw.githubusercontent.com/BioTender-max/awesome-bio-agent-skills/main/skills/{archive_path}/SKILL.md"
skill_md = urllib.request.urlopen(url, timeout=30).read().decode()
print(skill_md)   # read it, then follow its instructions

Some skills ship supporting files (scripts/, references/); list a folder via the GitHub API or clone it (Step 3) if you need them.

Step 3 — Offline / bulk / private fallback (clone)

If raw fetch is unavailable, or you want all supporting files, clone and copy the skill folder:

Standard (international):

git clone https://github.com/BioTender-max/awesome-bio-agent-skills.git
cp -r awesome-bio-agent-skills/skills/<archive_path>/ /path/to/your/agent/skills/

China-accelerated (ghfast.top mirror) — prepend the mirror to the GitHub URL; works for cloning this or any source repo (https://ghfast.top/https://github.com/<owner>/<repo>.git):

git clone https://ghfast.top/https://github.com/BioTender-max/awesome-bio-agent-skills.git
cp -r awesome-bio-agent-skills/skills/<archive_path>/ /path/to/your/agent/skills/

Categories (15) — counts and where to browse

Filter searches with the category key (left column). README section links jump to the browsable table for that category.

Source repositories (20)

Skills are aggregated and deduplicated from these repositories. Use the source_repo key to filter the index. (bio-agent-skills-hub is this self-referential hub entry.)

Example interactions

User: "I need a GWAS pipeline." Agent: search_skills("gwas") → top hit bio-workflows-gwas-pipeline (genomics, bioskills/gwas-pipeline); also gwas-pipeline (clawbio/gwas-pipeline) and plink2-gwas-analysis (sciagent/plink2-gwas-analysis). Fetch https://raw.githubusercontent.com/BioTender-max/awesome-bio-agent-skills/main/skills/bioskills/gwas-pipeline/SKILL.md and follow it. Browse more: https://github.com/BioTender-max/awesome-bio-agent-skills#genomics

User: "Preprocess and cluster my scRNA-seq data." Agent: search_skills("scrna preprocessing", category="single-cell") → scrna-preprocessing-clustering (bioclaw/scrna-preprocessing-clustering). Fetch its SKILL.md and follow it. 144 single-cell skills total: https://github.com/BioTender-max/awesome-bio-agent-skills#single-cell-analysis

User: "Design a protein binder backbone." Agent: search_skills("rfdiffusion") → rfdiffusion (protein-design, adaptyv/rfdiffusion). Related: proteinmpnn (adaptyv/proteinmpnn), bindcraft (adaptyv/bindcraft). Fetch the SKILL.md for the chosen tool. Browse: https://github.com/BioTender-max/awesome-bio-agent-skills#protein-design

User: "Call variants from long reads." Agent: search_skills("variant calling", category="genomics") → bio-long-read-sequencing-clair3-variants (bioskills/clair3-variants); for short reads see bio-variant-calling-deepvariant (bioskills/deepvariant). Fetch and follow.

Notes

• The index (bioskill_index_v3.csv) is the single source of truth for what exists and where; always resolve archive_path from it rather than guessing folder names.
• Counts above (1,676 skills / 20 sources / 15 categories) match the repository README and index.
• Content fetch requires network access and a public repository; use the clone fallback otherwise.