@galaxy-foundry/nfcore-tool-lab

Deterministic preparation of an already converted nf-core Galaxy tool for the experimental tools-iwc-lab repository. No model, checkout, network access, or GitHub credentials are needed. This does not convert Nextflow, open a PR, or deploy to a Tool Shed.

CLI

After publication, install with npm install -g @galaxy-foundry/nfcore-tool-lab, or run npx --package @galaxy-foundry/nfcore-tool-lab nfcore-tool-lab prepare .... In a Foundry checkout, build and run without publishing:

pnpm --filter @galaxy-foundry/nfcore-tool-lab build
node packages/nfcore-tool-lab/dist/bin/nfcore-tool-lab.js prepare \
  --input ./conversion \
  --output ./staging/seqkit/stats \
  --metadata ./lab-metadata.json

The input contains tool.xml, macros.xml, and the converter's _provenance.yml. Use --tool-file seqkit_stats.xml for an older explicitly named wrapper. Metadata is caller-supplied JSON, not inferred from the tool name:

{
  "description": "Compute sequence statistics",
  "categories": ["Sequence Analysis"],
  "homepage_url": "https://bioinf.shenwei.me/seqkit/"
}

All three fields are required; unknown fields are rejected. Choose real Tool Shed categories and the software's HTTPS homepage. The command does not verify these claims against upstream services.

Use --dry-run to validate inputs and print the preparation record without writing anything. Repeat --asset test-data or --asset helper.py to copy additional relative files/directories. Only the three required files and explicitly selected assets are copied, never the entire run workspace. Hidden paths, symbolic links, non-regular files, traversal, duplicate assets, and generated-filename collisions are rejected. Selecting assets is a trusted caller decision: the command cannot determine whether arbitrary selected file contents contain secrets. Executable assets retain executable permissions.

Macro imports must use safe relative filenames and resolve to a file in the prepared package. Missing imports fail with an explicit --asset suggestion; the command never automatically copies an imported file. Transitive imports are checked too, relative to the tool directory as in Galaxy (not the importing file's directory). Empty or unsafe paths, malformed imported XML, and import cycles are rejected, including during dry runs. Comments, command CDATA, and literal <import> elements nested in macro templates are not file imports.

The output directory must be new and outside the input tree, including through symlinked parent paths. No in-place update or overwrite mode is provided.

Preparation contract

Source identity comes from _provenance.yml.nfcore_source: modules_repo must be nf-core/modules; module_path must begin with modules/nf-core/ and contain only lowercase alphanumeric/underscore segments starting with an alphanumeric; both git_sha and test_datasets_sha must be full lowercase 40-character Git SHAs. The original generated converter identity and overrides must be present. Its cast_artifact_sha may be SHA-256 or YAML null; preparation neither guesses a missing bundle identity nor independently verifies it against a conversion run.

For modules/nf-core/seqkit/stats:

• Destination: galaxyproject/tools-iwc-lab:tool_collections/nf_core_modules/seqkit/stats
• XML tool ID and Tool Shed repository name: nfcore_compat_seqkit_stats
• XML tool name: original name plus (Nextflow Module Automated Conversion)
• Tool Shed owner: iwc-lab, never iuc

The two root attribute values are patched without XML serialization. Everything else in the wrapper, including CDATA, command indentation, line endings, profile, and version, stays byte-for-byte unchanged. XML must be well formed; DOCTYPEs and custom entities are unsupported. Naming is idempotent, and an already prefixed tool ID must match the source module. The patched XML is re-parsed to verify its ID and name match the preparation record.

Outputs are tool.xml, unchanged macros.xml and _provenance.yml, selected assets, .shed.yml, README.md, and _publication.json. The publication record has schema version 1, the preparation npm package/version, pinned source, destination, identity edits, and input/output content SHA-256 maps. Its own hash is excluded to avoid self-reference. It has no timestamp or random ID, so equal inputs with the same preparation version produce equal output bytes. Re-preparing an already prepared directory preserves naming but records that directory as the new input; it does not promise identical preparation records across that chain.

API

import { prepareLabTool } from "@galaxy-foundry/nfcore-tool-lab";

const record = prepareLabTool({
  inputDir: "./conversion",
  outputDir: "./staging/seqkit/stats",
  metadata: {
    description: "Compute sequence statistics",
    categories: ["Sequence Analysis"],
    homepage_url: "https://bioinf.shenwei.me/seqkit/",
  },
  assets: ["test-data"],
  dryRun: false,
});

The API returns the same typed preparation record printed as JSON by the CLI. CLI failures exit nonzero and write diagnostics to stderr; API failures throw.

Publication is a later step

validation.status is always not_run. This command does not certify licensing, test coverage, upstream authenticity, or wrapper behavior, and .shed.yml's type: unrestricted is repository metadata, not a software-license judgment. Preserve any coverage gaps in the conversion provenance and review them explicitly.

Before creating a PR, the publishing harness must run final-package Planemo lint, Tool Shed metadata checks, and real tests; verify licensing and conversion bundle identity; and check repository-wide tool/Shed-name collisions. Flattening / to _ can give two different module paths the same ID, so collision detection cannot be settled from one input directory. The harness must create a scoped branch/PR, disclose AI generation, and prove destination CI discovers this nested repository. Packaging never enables deployment or needs publishing credentials.