awesome-genomic-skills
A discovery index above the genomics skill libraries (it catalogs them) — its framing, strengths, and where it sits on our values.
github.com/GoekeLab/awesome-genomic-skills — "A curated list of skills and MCP servers for working with AI coding agents in genomics and bioinformatics." Maintained by GoekeLab; CC0-1.0. A curated markdown index — not a skill library — that catalogs ~50+ entries across five sections: genomics agent-skill collections (science-skills, openai life-science-research, anthropics/life-sciences, AWS Kiro, ClawBio, SciAgent-skills, bioSkills, operon), MCP servers (biomcp, gget-mcp, Seqera, knowledgebase-mcp, registries), benchmarks (BioAgent Bench, BiomniBench, BixBench, CompBioBench, SkillsBench), and general agent-skill lists, subdivided by runtime (Claude Code, Copilot, Codex, Cursor, Gemini CLI). Verified 2026-06-27 against the repo README.
Its framing
awesome-genomic-skills is a discovery index for the genomics agent-skills ecosystem: a hand-curated, CC0 "awesome list" that points to where skills, MCP servers, and benchmarks live, with a short description and developer attribution per entry. It hosts no SKILL.md files of its own — every entry is a link out to an external project (it catalogs bioSkills, ClawBio, operon, science-skills, and others side by side). Its value is breadth and orientation: a single reading surface from which a practitioner can find and compare the libraries, servers, and evaluations now appearing across the field, organized by domain and by which coding agent they target.
Strengths relative to our approach
It leads the Foundry cleanly on ecosystem coverage and on being a thing a human actually reads:
- Ecosystem breadth no single library matches. It is the map of the territory — skill libraries, MCP servers, and benchmarks from DeepMind, OpenAI, Anthropic, AWS, GPTomics, ClawBio, and academic labs in one place, cross-referenced by runtime. The Foundry is one library on this map; awesome-genomic-skills is the map.
- A genuinely human-facing surface. Unlike the file-directory neighbors (bioSkills, ClawBio), this is authored to be read — curated prose, grouped by domain and agent, with per-entry attribution. It is the closest of the surveyed projects to "a destination a human opens to orient," even if it orients rather than teaches.
- Open, low-friction contribution. CC0-1.0 and a flat markdown list make it trivially forkable, citable, and PR-extendable — and a natural place for cast skills to be listed.
- Benchmark-aware. It foregrounds the evaluation layer (SkillsBench's "+16.2 pts" pass-rate claim for curated skills, BiomniBench, BixBench, CompBioBench), tracking how the field measures agent competence — a layer most skill libraries omit entirely.
Where it sits on our values
| Value | awesome-genomic-skills |
|---|---|
| Produces skills | No (by design) — purely curatorial; it links to others' skill libraries and authors none itself. Neutral, not a flaw: it is an index, a different kind of object than a skill producer. |
| Progressive disclosure | No — a flat curated list (category → entry → outbound link). There is a category/entry hierarchy, but no name→body→on-demand-reference disclosure of any skill, because it contains no skills. |
| Traceability | Strong at the index level — every entry is attributed to its developer and hyperlinked to its source, so a human can follow each pointer to origin. But the links are human-followable markdown, not machine-resolved cross-references, and traceability stops at the boundary: it vouches for where a skill lives, not for what any skill cites internally. |
| CLI instructions derived from CLI docs | N/A — wraps no CLI tools; it is a catalog, not an analysis or tool-invoking artifact. |
| Portability | No skills produced — it ships no artifacts to port. It does index skills across many runtimes (Claude Code, Copilot, Codex, Cursor, Gemini CLI), but cross-runtime breadth here is a property of the things it lists, not of itself. |
| Human scrutiny | The list itself, yes — readable CC0 markdown, PR-contributable. But it presents pointers and one-line descriptions, not skill internals: a reader scrutinizes the catalog's curation, not the data inside any skill. |
| Knowledge-base backed | No — the curated list is the artifact; there is no inspectable knowledge base beneath it, only outbound links to the projects it indexes. |
Where it diverges from the Foundry
awesome-genomic-skills and the Foundry sit on different rungs of the same ladder: it is a discovery index above the skill libraries, while the Foundry is a skill producer with a knowledge base beneath its artifacts — and it catalogs neighbors like bioSkills and ClawBio that the Foundry positions against directly. The shared axis, credited as a similarity, is that both present a human-readable surface in a field of file-directories; but the index orients (here is where skills live) rather than teaches (here is how a method works and why it is valid), and it foregrounds links, not a progressive-disclosure learning surface with provenance. By its own framing it makes no claim on statistical rigor — it does not mandate null/permutation calibration, FDR control, negative controls, or a referee gate, because catalogs do not run analyses; that empirical-honesty layer is simply orthogonal to what an index does. The crux: it is the field's table of contents, the natural place a cast Foundry skill would be listed — not a competitor to what the Foundry casts.
Sources
- https://github.com/GoekeLab/awesome-genomic-skills — README: self-description, CC0-1.0 license, GoekeLab maintainer, the five curated sections, ~50+ indexed entries, per-entry developer attribution, runtime subdivision, and the absence of any hosted skill artifacts.