ASB Skill Collections

Evidence-grounded skills and software-tool records for scientific AI agents,
distilled from published method papers and their code by the AgenticScienceBuilder pipeline.

A skill is a short procedure that an AI agent can read and follow. The AgenticScienceBuilder (ASB) pipeline writes each one from a method paper (54 of the 566 are preprints) and, where one exists, its public code repository, anchors it to verbatim evidence from those sources (for repo-oa papers, the repository's own documentation; CONTENT_POLICY.md), annotates it with EDAM terms, and passes it through a licence and PII gate before release. This repository is the public registry of those collections: the released metabolomics collection, its per-technique packs, the install and search tooling, and the governance of the community that curates them. The project started at the Dagstuhl Seminar Computational Metabolomics (26181, 26 to 30 April 2026).

This release

collections/metabolomics/v2, release metabolomics-v0.2.0 (preliminary): 5,859 skills across 909 tools, distilled from 566 papers, plus 21 composite workflow super-skills that chain the atomic skills into end-to-end pipelines (workflows/). Coverage is mostly LC-MS/MS, with LC-MS, GC-MS, mass-spectrometry imaging, ion mobility and lipidomics, some NMR, and multi-omics, statistics and pathway tools.

Layer	Contents	State
ASB-Skills	evidence-grounded procedural skills	released
ASB-Workflows	composite end-to-end pipeline super-skills	released
ASB-Tools	deduplicated software-tool records with EDAM terms and DOIs	released
ASB-Benchmark	per-paper tasks and claim-retrieval test sets	later
ASB-Capsules	raw per-paper pipeline outputs, for full traceability	later

Install

Claude Code (plugin marketplace, public GitHub route):

/plugin marketplace add HolobiomicsLab/asb-skill-collections
/plugin install metabolomics@asb-skill-collections

This route was verified at CLI level on 2026-10-30 with Claude Code 2.1.304: claude plugin marketplace add HolobiomicsLab/asb-skill-collections and claude plugin install metabolomics@asb-skill-collections both exited 0. Loading the skills in a live session was not exercised. The marketplace keeps a clone of this repository (286 MB) beside the plugin cache (72 MB), about 358 MB on disk. From a local checkout, claude plugin marketplace add "$PWD" then claude plugin install metabolomics@asb-skill-collections installs the same plugin without that clone (USAGE §1). Per-technique packs such as metabolomics-lc-ms are lighter subsets (packs/metabolomics): metabolomics-lc-ms was installed the same way at CLI level (29 MB plugin cache), and the other seven packs were built but not installed.

Other agent runtimes (asbb CLI from a checkout; Python 3.12 or later and PyYAML, in a virtual environment):

git clone https://github.com/HolobiomicsLab/asb-skill-collections
cd asb-skill-collections
uv venv -p 3.12 --allow-existing && source .venv/bin/activate && uv pip install pyyaml
python3 -m asb_skill_collections.asbb_cli install metabolomics --runtime agents   # ~/.agents/skills (Codex, Copilot CLI, Gemini CLI)
python3 -m asb_skill_collections.asbb_cli install --list-runtimes                 # agents (verified), codex, copilot, gemini, claude, cursor, cline, vscode-copilot, --dest DIR (experimental)

Without uv, python3.12 -m venv .venv && . .venv/bin/activate && python -m pip install pyyaml prepares the same environment. On macOS the environment is needed: Homebrew's Python refuses a system-wide pip install, and Apple's /usr/bin/python3 is older than 3.12. Each install copies the unit into a content-addressed snapshot that survives a moved checkout; rerun install to refresh (usage guide: commands, outputs, uninstall). Any other agent can read the Markdown and JSON under collections/metabolomics/v2/ directly (AGENTS.md). Not on PyPI in v0: an index install would fetch the CLI without the corpus it reads. Verified routes and dates: docs/STATUS.md.

Choose what to install

Install the whole collection, a technique pack, a selection of skills (asbb install --skills, --skills-file or --query), or fetch chosen skills without a checkout (asbb fetch). Commands, outputs and refusals: docs/partial-install.md.

Install	Technique tag	Leaves	On disk (du)
metabolomics, the full collection	all	5,859	72 MiB
metabolomics-lc-ms	LC-MS	2,617	29 MiB
metabolomics-ms-generic	mass-spectrometry	804	8.5 MiB
metabolomics-ion-mobility	ion-mobility-MS	385	4.3 MiB
metabolomics-gc-ms	GC-MS	367	4.0 MiB
metabolomics-ms-imaging	MS-imaging	291	3.2 MiB
metabolomics-nmr	NMR	276	2.9 MiB
metabolomics-ce-ms	CE-MS	113	1.4 MiB
metabolomics-direct-infusion	direct-infusion-MS	97	1.1 MiB
a selection, asbb install --skills a,b,c --collection metabolomics/v2 --runtime agents	chosen slugs	10 in the example	188 KiB
a fetch, asbb fetch a,b --collection metabolomics/v2 --dest DIR	chosen slugs	10 in the example	112 KiB

The first nine rows install by name: /plugin install <name>@asb-skill-collections in Claude Code, where the public marketplace also keeps its 286 MB clone of this repository, or python3 -m asb_skill_collections.asbb_cli install <name> --runtime agents. Their sizes are du -sk of the built unit in this repository on 2026-10-31; the selection and fetch rows are the 10-skill example of docs/partial-install.md, and a fetch also downloads the 5.2 MiB (5,430,046 bytes) skills_index.json. du counts allocated blocks, so the full collection that takes 72 MiB here is the 61 MB of file content that docs/partial-install.md reports.

Packs overlap: a skill tagged with several techniques is in several packs, so install the full collection or a few packs, not both. The 1,520 technique-agnostic skills (statistics, pathway and generic data-handling procedures) carry no technique tag and are in no pack; they come with the full collection, a selection or a fetch.

Use: search, apply, ground

Search a checkout offline, with no API key:

uv venv -p 3.12 --allow-existing && source .venv/bin/activate
uv pip install -e .            # in a python3.12 -m venv environment: python -m pip install -e .
asbb search "untargeted LC-MS/MS annotation" --collection metabolomics --target workflows
asbb get untargeted-lcmsms-annotation --collection metabolomics/v2 --target workflows
asbb search --list-collections

Without installing, python3 -m asb_skill_collections.asbb_cli runs the same commands. The CLI, the router, the MCP server and the collection-level search scripts share one versioned selector; each hit reports its score, matched fields and qualified slug. How offline skill selection works documents the rule, its tie handling and its 22-query measurement.

Apply. Read the skill's SKILL.md and follow the procedure. Tool requirements are in its frontmatter and in USAGE §0.

Ground. To check an exact parameter, threshold or claim (the difference between "the skill says about 5 ppm" and "the paper specifies 5 ppm"), ground the skill against its source paper. Grounding is optional, ships inside every plugin and pack, and uses Perspicacité, Holobiomics Lab's local-first literature RAG engine, with two backends:

• kb: RAG over the paper's full text and supplementary information. The per-paper KB (asb-paper-<doi>) is created on first use through the Perspicacité HTTP API; no MCP server is required. Needs Perspicacité at PERSPICACITE_BASE (default http://127.0.0.1:8000).
• local: no server. Clones the skill's source repository and fetches the open-access paper where possible. Needs git and network.

In Claude Code, run /ground on the skill in play; elsewhere, call the bundled bin/perspicacite_kb_bind.py (prepare, query, local). Full guide: USAGE §4.

What is in the collection

_{The LC-MS view of the collection (the metabolomics-lc-ms pack). The full release covers all techniques.}

File	Contents
leaves/<slug>/SKILL.md	one evidence-grounded skill each (frontmatter: EDAM IRIs, derived_from DOIs, tools, attribution; verbatim evidence lines in the body). Shipped as data and read on demand, so installing does not load the corpus into the session's context.
skills/*/SKILL.md	three advertised entries: the router, the collection and licence policy, the contribution channel
workflows/<slug>/SKILL.md	composite workflows and their router, workflows/_workflow_router/SKILL.md
tools/<slug>.yaml	deduplicated software-tool records with EDAM terms and source DOIs
skills_index.json, tools_index.json	machine search indexes (EDAM IRI, tool name, keyword)
kb_bundle.json	skill to source-paper KB slugs and repo_urls (the grounding map)
bin/perspicacite_kb_bind.py, commands/ground.md, GROUNDING.md	packaged grounding: the binder, the /ground command and a how-to
commands/*.md	the four plugin commands: /ground; /propose-skill, which stages a community skill proposal; /synthesize-meta-skill, which stages a workflow super-skill proposal; /claim-skill, run by a maintainer to turn a tool author's claim into a proposal that credits the author
collection.yaml, corpus.yaml	collection record; per-paper access basis (repo-oa)
CITATION.cff, PROVENANCE.md, gate_report.json	citation, how it was generated, release-gate verdict

The default entry point is skills/_router/SKILL.md. Technique and EDAM-topic breakdowns and the selection criteria are in ABOUT.md.

How it was generated

Skills come from method papers anchored to the computational metabolomics review series (SOURCES.md: sources and inclusion criteria; CONTENT_POLICY.md: legal and open-access policy). PROVENANCE.md records the exact build command and model routing: Claude Opus 4.8 for outline and card revision, Claude Haiku 4.5 for the other stages, OpenAI text-embedding-3-large and -small for embeddings, each build recorded in build_manifest.json.

Repository family

Repository	Role	Access
asb-skill-collections (this one)	registry, release tooling, governance, released metabolomics collection	public
AgenticScienceBuilder	the pipeline and schema that generate the collections	private, to follow
asb-skills-<domain>	one repository per domain, same collections/<domain>/<version>/ and packs/<domain>/<technique>/ layout	private, to follow
indicium	standard for a typed scientific claim bound to the evidence that grounds it	private, to follow
indicium-adapters, indicium-adapters-<domain>	per-domain adapters that enrich indicium claims with ontology terms	private, to follow
Perspicacite-AI	literature RAG engine used for grounding	public
Mimosa-AI	multi-agent framework for autonomous scientific research	public

Domain and adapter repositories exist for metabolomics and nine further domains (agentic AI, astronomy, epigenomics, genomics, metabarcoding, metagenomics, multi-omics, proteomics, transcriptomics) and will be published with their releases. Until then the metabolomics collection is served from here; collections/ also holds staged epigenomics/v1, transcriptomics/v1 and metabolomics/v1, not part of this release.

Attribution and citation

Cite the collection and the original paper behind each skill you use (attribution.original_doi in its SKILL.md).

• Concept DOI, for all versions: 10.5281/zenodo.20794026.
• Version DOI of the deposited version, metabolomics-v0.1.0 (22 June 2026): 10.5281/zenodo.20794027. The metabolomics-v0.2.0 content of this repository is not yet deposited.
• Authors. The v2 deposit (CITATION.cff): AgenticScienceBuilder Community, Louis-Félix Nothias, HolobiomicsLab.cnrs.fr, MetaboLinkAI.net. From the next versioned release on (root CITATION.cff): Louis-Félix Nothias and the AgenticScienceBuilder AI framework, joined by curators (call for curators).
• Per-skill roles (attribution: in each SKILL.md): generator (the ASB pipeline), curators (none yet), promoter (Louis-Félix Nothias), sponsor (CNRS and Université Côte d'Azur, who paid the API cost), original_doi.

Governance and licensing

Every skill carries a license_tier (in skills_index.json and in its SKILL.md frontmatter) that answers what may I do with the underlying tool?

Tier	Meaning
open	commercial use allowed (MIT, Apache-2.0, GPL, CC-BY, ...)
noncommercial	academic or noncommercial use only; confirm permitted use before applying the skill
restricted	licence established and it constrains reuse; check it before commercial use or redistribution
unknown	no tool-level licence evidence; describing and using are unaffected; never redistribute the tool's code

The search tools do not filter by tier. The router tells the agent to prefer open skills when no tier is requested, and to get confirmation of a permitted (academic or noncommercial) use before applying a noncommercial one; the asb-metabolomics meta-skill carries the same acknowledgement step. A technique pack's router carries the noncommercial confirmation inline but no open-first default (LICENSE_TIERS.md). Code is under Apache-2.0, collection content under CC-BY-4.0; verbatim quotations (evidence_spans) are minimal and attributed, and a rights holder may request removal (LICENSING.md). Other policies are in governance/, starting with MAINTAINERS.md and COI_POLICY.md.

Contributing

• A skill was wrong, stale or missing. The asb-contribute skill, shipped with the plugin, turns it into a redacted report you approve before anything is filed; from a checkout, scripts/make_improvement_report.py prints an anonymised issue body, and with --print-url a pre-filled issue URL.
• Propose or annotate a paper, a collection or a workflow: CONTRIBUTING.md.
• Credit. Merged contributions count on the contributor leaderboard and can lead to co-authorship of the next release DOI (CONTRIBUTION_LOOP.md).
• Before packaging, run python -m scripts.unit_closure from the repository root: it compares each shipped unit's indexes with its leaves and checks declared metadata.helper files.

Status

• Verified for this release: the full metabolomics collection as a managed install, Claude Code from a local-checkout marketplace, and the checkout CLI.
• Verified at CLI level on 2026-10-30 (Claude Code 2.1.304), in-session load not exercised: Claude Code from the public GitHub marketplace (about 358 MB on disk) and the metabolomics-lc-ms pack.
• Pending or experimental: the other seven packs, the web UI, runtime adapters other than agents, the MCP server. Not yet published: PyPI, the documentation site, the HuggingFace mirror, the benchmark and the capsules.
• The w3id.org/holobiomicslab/ and w3id.org/asb identifiers resolve. Dates, measurements and detail per surface: docs/STATUS.md.

Acknowledgements

We thank the participants and organisers of Dagstuhl Seminar 26181, Computational Metabolomics, where this collaboration began. API generation costs for this collection were sponsored by CNRS and Université Côte d'Azur. Maintained by Holobiomics Lab (CNRS, Université Côte d'Azur).