Foundry
Home Schema

Nextflow pipeline summary

JSON Schema for the structured summary emitted by the summarize-nextflow Mold.

draft source/nextflow
Revised
2026-09-15
Rev
13
Related
schema summary-nextflow @galaxy-foundry/summarize-nextflow @ 0.3.0 upstream ↗

Nextflow Pipeline Summary

Structured per-source summary emitted by the summarize-nextflow Mold. Per-source schema by design — paper, Nextflow, and CWL each have their own summary shape; downstream Molds (data flow, templates, tool wrappers) consume any source's summary and handle the polymorphism. Field names mirror gxy-sketches' SketchSource / ToolSpec / TestDataRef / ExpectedOutputRef where parity exists; see content/research/gxy-sketches-alignment/index.md.

31 sections. Anchor links per section (e.g. #root) are stable.

Channel

Channel

One top-level channel constructed in the workflow. Sources include `Channel.fromPath`, `Channel.fromFilePairs`, `samplesheetToList`-driven `Channel.fromList`, `splitCsv`, `file`/`files`, and plain `params.*` references.

field type req description
construct string ✓ Classifies the channel's primary materialization factory or shape-determining operator. Selection precedence: (1) `samplesheetToList` when the chain contains `samplesheetToList(...)` (typically wrapped in `Channel.fromList`); (2) `splitCsv` when the chain ends in `.splitCsv(header: true)` over a path; (3) otherwise the outermost factory (`Channel.fromPath` → `fromPath`, `Channel.fromFilePairs` → `fromFilePairs`, `file(...)` → `file`, etc.); (4) `other` for unrecognized constructions. The verbatim `source` retains the full chain — `construct` is the typed lookup the converter would otherwise re-parse.
from_param string | null ✓ Foreign key into `params[].name` when the construction expression directly references `params.X` (e.g. `Channel.fromPath(params.reads)`, `samplesheetToList(params.input, ...)`, `file(params.fasta)`). Null for literal-glob construction, expressions that compose multiple params, and channels derived from other channels. v1 resolution is direct-only; one-hop Groovy bindings (`def reads = params.reads; Channel.fromPath(reads)`) are not chased — see galaxyproject/foundry#211.
name string ✓
required_runtime boolean ✓ True when the construction chain ends in `.ifEmpty { error ... }` (or an equivalent imperative emptiness-throw guard). Captures runtime requiredness even when the param's nf-schema entry does not mark it required. Combine with `params[].required` and any imperative pre-construction `error` checks for the full requiredness picture.
shape string ✓ String-encoded channel shape; same convention as ChannelIO.
source string ✓ Verbatim channel-construction expression.

ChannelIO

ChannelIO

One declared input or output channel of a process. Channel shape is a string, not a structured type — `tuple(meta, [path,path])` is enough for downstream Molds and avoids a research project on NF channel typing.

field type req description
name string ✓
shape string ✓ String-encoded channel shape, e.g. `tuple(meta, [path,path])`, `path`, `val(integer)`.
description string
topic string | null Nextflow channel-topic name when this output is bound to a topic (e.g. `versions` for the standard nf-core version-aggregation topic). Topics are a Nextflow 24+ feature; the module-level `topic: <name>` annotation on a process output emits to a global named channel that any consumer can `channel.topic('<name>')` to read. Null for non-topic outputs and for all inputs.

Conditional

Conditional

One workflow-level conditional that gates a subgraph.

field type req description
affects string[] ✓ Process or subworkflow names whose execution is gated by this conditional.
branch string ✓ Which side of the conditional this entry describes. `default` is the truthy branch; `alternate` is the else branch.
guard string ✓ Verbatim guard expression, e.g. `params.skip_alignment`.

Edge

Edge

One edge in the workflow's call graph. The deterministic parser records the literal operator chain in `via`; reconciliation of the source and target shapes is the LLM step.

field type req description
from string ✓ Channel name or `<PROCESS>.out.<chan>` reference.
to string ✓ Process or subworkflow name receiving this channel.
notes string LLM-emitted note when the reconciliation is low-confidence (e.g. deeply nested closures).
via string[] Operators on the path between source and target, in order (e.g. `["map", "join", "groupTuple"]`). Empty for a direct edge.

Evidence

Evidence

Shared provenance block for source-detected facts. `source_path` localizes the finding; `confidence` is a coarse self-assessment; `evidence[]` carries free-form snippets (matched lines, normalized expressions, detector notes). Reused by ReferenceAsset and ReferenceRebuildRule.

field type req description
confidence string ✓ Detector self-assessment. `high`: all signals match the expected idiom; `medium`: some signals match but the guard or builder is partial / mixes non-param locals; `low`: heuristic match, downstream should treat as a hint.
evidence string[] ✓ Free-form evidence snippets (matched source lines, normalized expressions, detector notes). Empty when none.
source_path string | null ✓ Repo-relative path of the source file where the finding was detected (e.g. `subworkflows/local/prepare_genome/main.nf`, `nextflow.config`). Null when the finding spans files or cannot be localized.

ExpectedOutputRef

ExpectedOutputRef

One expected output. Field names mirror gxy-sketches' ExpectedOutputRef verbatim. At least one of path / url / assertions is required.

field type req description
role string ✓
assertions string[] Assertion strings. Simple equality / regex / contains-string checks are preserved verbatim; complex Groovy assertions are summarized to prose with a `warnings[]` flag in the parent Summary.
description string | null
kind string | null Coarse output kind: `report`, `tabular`, `image`, `archive`, `sequence`, `other`.
path string | null
url string | null

InvocationBinding

InvocationBinding

Binding of one positional argument to a callee `take:` name.

field type req description
argument string ✓ Verbatim caller-side argument expression.
take string ✓ Callee `take:` name (FK into the callee subworkflow's `inputs[].name`).

ModuleMeta

ModuleMeta

Normalized subset of nf-core module `meta.yml`, captured next to a process so module-scoped downstream Molds can consume `processes[i]` without reading the full pipeline summary.

field type req description
authors string[] ✓
input → ModuleMetaEntry[] ✓
keywords string[] ✓
maintainers string[] ✓
output → ModuleMetaEntry[] ✓
tools → ModuleMetaEntry[] ✓
description string

ModuleMetaEntry

ModuleMetaEntry

One named entry from a module `meta.yml` section, normalized from nf-core's single-key YAML maps into `{ name, ...fields }` objects.

field type req description
name string ✓
description string
documentation string
doi string
homepage string
identifier string
licence string[]
pattern string
tool_dev_url string
type string

NfTest

NfTest

One `tests/*.nf.test` file. nf-core templates use a near-uniform shape: a `nextflow_pipeline { test("<desc>") { when { params { ... } } then { assertAll(...) } } }` block, with one test() per profile, asserting `workflow.success` plus a snapshot match. This shape captures the data so downstream test-conversion molds (e.g. nextflow-test-to-target-tests) don't have to re-parse Groovy.

field type req description
assert_workflow_success boolean ✓ Whether the test asserts `workflow.success`. Almost always true for nf-core templates.
execution_mode string ✓ Whether the test requests real execution, stub execution, or mixes both.
name string ✓ Description string passed to `test("...")`. Often duplicates the profile name.
params_overrides object ✓ The `when { params { ... } }` block as a key→value map. Most templates set only `outdir`; pipeline-specific overrides land here.
path string ✓ Repo-relative path of the .nf.test file (e.g. `tests/dfast.nf.test`).
profiles string[] ✓ Profile name(s) the test runs under, from the file-level `profile "<name>"` declaration and/or per-test config overrides. Usually a single profile.
prose_assertions string[] ✓ Other assertions (regex matches, count checks, content equality on specific files) summarized to prose strings. Empty for snapshot-only tests, which is the common nf-core case.
snapshot → SnapshotFixture | null ✓ Structured representation of the `assert snapshot(...).match()` clause when present; null when the test uses no snapshot.

Param

Param

One pipeline parameter. Sourced from `nextflow.config`'s `params { ... }` block, augmented from `nextflow_schema.json` when present (nf-core).

field type req description
name string ✓
required boolean ✓
type string ✓ JSON Schema-style type when nextflow_schema.json supplies it (string, integer, number, boolean, path); free-form when inferred from a default value.
default any Default value verbatim from the source. May be null.
description string
enum any[] Allowed values when nextflow_schema.json declares them.
fa_icon string | null Font Awesome icon class from the parent `$defs` section's `fa_icon` (e.g. `fas fa-terminal`). Optional UI hint; null when undeclared.
format string | null JSON Schema `format` keyword from nextflow_schema.json. nf-schema vocabulary: `file-path`, `directory-path`, `path`, `file-path-pattern`. Distinguishes path-typed params (datasets/collections) from plain strings without re-reading the source schema. Null when undeclared.
hidden boolean | null nf-schema `hidden` keyword. True for CLI-plumbing params (e.g. `validate_params`, `pipelines_testdata_base_path`, `version`) that shouldn't surface in user-facing target interfaces. Null when undeclared.
mimetype string | null nf-schema `mimetype` keyword (e.g. `text/csv`, `text/plain`, `application/gzip`). Useful for seeding Galaxy `format` on path-typed params. Null when undeclared.
schema_group string | null Human-readable group label from the parent nextflow_schema.json `$defs` section's `title` (e.g. `Input/output options`, `Reference genome options`). Preserves nf-schema sectioning so target interfaces can group inputs without re-parsing the source. Null when the param is not under a titled section.
source_expression string | null Verbatim right-hand-side expression when the param was assigned in Groovy (e.g. `getGenomeAttribute('fasta_fai')`). Null when the param comes from a JSON Schema entry only.
source_kind string | null How this param entry was derived. `nextflow_schema`: declared in `nextflow_schema.json`. `params_block`: declared (or defaulted) in `nextflow.config`'s `params { ... }` block. `getGenomeAttribute`: synthesized from `params.X = getGenomeAttribute('X')` assignments in `nextflow.config` (nf-core key-expanded bundle pattern). `computed`: derived from another non-getGenomeAttribute expression. Null when provenance was not classified.
source_path string | null Repo-relative path of the source file where the assignment lives (e.g. `nextflow.config`, `conf/igenomes.config`). Null when undefined or when the param comes only from `nextflow_schema.json`.

Process

Process

One Nextflow `process { ... }` block.

field type req description
inputs → ChannelIO[] ✓
meta → ModuleMeta | null ✓ Normalized `meta.yml` from the module directory when present. Null for local/ad-hoc processes without module metadata.
module_path string ✓ Relative path from the pipeline root to the file declaring the process.
module_tests → NfTest[] ✓ Module-scoped nf-test files under the module's `tests/` directory. Empty for local/ad-hoc modules without unit tests.
name string ✓ Canonical process name (uppercase NF convention) as declared by the `process <NAME>` line in `module_path`.
outputs → ChannelIO[] ✓
aliases string[] Alias names this process is imported as in workflow scopes via `include { <NAME> as <ALIAS> }`. Real nf-core pipelines re-import a single module multiple times under different aliases to invoke it with distinct runtime args (e.g. `MINIMAP2_ALIGN as MINIMAP2_CONSENSUS`, `MINIMAP2_ALIGN as MINIMAP2_POLISH`). Each alias appears in `workflow.edges[].from`/`.to` exactly as written; the canonical `name` does not appear in edges unless an unaliased import exists. Default `[]` when the process is only imported under its canonical name.
conda string | null Verbatim text of the process's `conda` directive (or null). Two common shapes: a literal `bioconda::<name>=<version>` (legacy) or a file reference `${moduleDir}/environment.yml` (modern nf-core convention). Either way the resolved bioconda spec lands in `tools[].bioconda`.
container string | null Verbatim text of the process's `container` directive (or null). Modern nf-core modules use a ternary expression `"${ workflow.containerEngine == 'singularity' && !task.ext.singularity_pull_docker_container ? '<singularity-uri>' : '<docker-uri>' }"` — keep that text intact here. The two resolved branches are split out into `tools[].docker` / `tools[].singularity` (and `tools[].wave` / `tools[].biocontainer` as appropriate).
in_subworkflows string[] Denormalized FK: names of `subworkflows[]` whose `calls[]` invoke this process (canonical name or any alias). Empty `[]` when the process is only invoked from the primary `workflow` body. Useful for grouping the flat process-tier DAG into a subworkflow-tier view client-side. Multi-entry when the same module is invoked from several subworkflows (typically via aliases).
publish_dir string | null `publishDir` value when the process publishes outputs (or null).
script_excerpt string | null Verbatim `script:`/`shell:`/`exec:` body, dedented and truncated at 4000 characters, or null when the process declares none. Deterministic. For ad-hoc pipelines with no `meta.yml` this is the only evidence of what a process does.
script_summary string One-line LLM-derived summary of what the `script:` body does. Free text. The deterministic CLI does not emit it — it emits `script_excerpt` instead, and the cast skill's LLM pass fills this in.
tool string | null Foreign key into `tools[].name` when the process runs a single recognizable tool; null for multi-tool or pure-Groovy processes.
when string | null Verbatim `when:` guard expression, or null if absent.

Profile

Profile

One entry from a `profiles { }` block, classified by role. A `test` kind supplies profile-level candidate evidence without claiming that the profile is the selected, cheapest, or most representative test case. Pipeline-level `test_candidates[]` entries derived from nf-test are the higher-fidelity candidate unit because they preserve per-test parameter overrides.

field type req description
includes string[] ✓ `includeConfig` targets named directly in the profile body, verbatim and unresolved (e.g. `conf/test_full.config`). Interpolated paths are recorded as written. The classifier follows these to reach the real settings; downstream consumers can use the config family as one ranking signal, but it does not by itself establish runtime cost or representative coverage.
kinds string[] ✓ The roles the profile body evidences. An array rather than a scalar because real profiles combine roles — 25 of the 371 corpus profiles select a container engine *and* an executor in one block (`CRG-CNAG/CalliNGS-NF`'s `cluster` sets `process.container`, `singularity.enabled`, and a `process { }` block with `executor = 'crg'`; seven of `biocorecrg/MOP2`'s eight profiles do the same). Values: `test` supplies input data matching the recognized test conventions — includes a `conf/test*.config`, or assigns `params.input` / `samplesheet` to a non-null value; `test_full` is this kind too, with its config family visible in `includes`. `container` selects a container or package-management engine (`docker.enabled`, `process.conda`, `process.arch`, ...). `executor` selects an executor, cloud backend, or `workDir`. `resources` tunes process directives only (`memory`, `cpus`, `time`, `withLabel:` selectors). `mode` assigns params that change pipeline behavior without supplying input data (`nf-core/rnaseq`'s `prokaryotic`, `nf-core/smrnaseq`'s protocol presets). `dev` is a developer-ergonomics profile (`debug`, `gitpod`). `unknown` means no classifying signal was found. Not every role is additive. `test`, `container`, `executor`, and `dev` accumulate. `mode` and `resources` are assigned **only when none of those matched**, because their signals are far weaker — nf-core's `docker` profile assigns `params.use_gpu` inside a ternary, and most test profiles set process resource caps, so accumulating them would label most of a corpus `mode`. A consumer that needs the suppressed evidence reads `signals[]`, where `params-assignment` and `process-directives` survive regardless. `test` is profile-level candidate evidence, not a final selection or a claim about cost or representative coverage. Prefer nf-test-derived `test_candidates[]` entries when present because they preserve per-test parameter overrides; use an input-bearing profile as a fallback when no pipeline-level nf-test case exists. Mode profiles may encode science-scope choices that target translation must surface, unlike execution-only container and executor profiles. `test` detection keys on the nf-core input conventions (`conf/test*.config`, `params.input`, `params.samplesheet`). All 103 corpus `test` profiles are in the 16 nf-core fixtures and none in the 10 ad-hoc ones — an ad-hoc pipeline naming its inputs `reads` / `genome` yields no `test` profile even where one exists in spirit. Absence of a `test` kind therefore means 'not detected', not 'none declared'.
name string ✓ Profile name as declared, i.e. the value passed to `-profile`.
signals string[] ✓ Evidence the classifier acted on, so a surprising `kinds` value is auditable without re-reading the pipeline config — and so evidence suppressed by the `mode` / `resources` fallback rule is still visible. Empty exactly when `kinds` is `["unknown"]`: nothing matched, or the `includeConfig` chain holding the settings was deeper than the resolver follows.
source_path string ✓ Repo-relative path of the config file whose `profiles { }` block declared this profile. Not always `nextflow.config` — `epi2me-labs/wf-human-variation` declares all six in `base.config`, `biocorecrg/MOP2` in `nextflow.global.config`.

ReferenceAsset

ReferenceAsset

Curated source-side view of one reference-data input the pipeline consumes. A foreign-key projection onto `params[].name` plus the metadata downstream classification Molds need. Target-agnostic: no Galaxy translation decisions are encoded here.

field type req description
asset_kind string ✓ Coarse asset classification (e.g. `fasta`, `fasta_index`, `sequence_dictionary`, `bwa_index`, `bwamem2_index`, `tabix_index`, `gtf`, `gff`, `bed`, `vcf`, `database`, `reference_sheet`, `other`). Free-form string; downstream Molds choose how to map onto target datatypes.
evidence → Evidence ✓
param string ✓ FK into `params[].name`.
required boolean ✓ True when the source pipeline treats this asset as required (no compute-if-missing branch, no optional guard). When false, see `reference_rebuilds[]` for any rebuild branch covering this param.
used_by string[] ✓ Names of processes / subworkflows that receive this param as an argument (via direct invocation binding or channel construction). Empty when no caller could be attributed.
format_hint string | null Format hint, often the nf-schema `format` keyword (`file-path`, `directory-path`, `path`, `file-path-pattern`) or a domain hint like `fai`, `dict`. Null when undetermined.
schema_group string | null Mirrors `params[<this>].schema_group`. Useful for grouping (`Reference genome options`).
source_expression string | null Mirrors `params[<this>].source_expression`. Duplicated for the same reason.
source_kind string | null Mirrors `params[<this>].source_kind`. Duplicated here so reference-data consumers don't have to dereference.

ReferenceRebuildRule

ReferenceRebuildRule

One compute-if-missing rebuild branch detected in a workflow or subworkflow body. Binds an asset param to the guard expression and builder process that produces it when absent. Source evidence only; in-tool-rebuild decisions live downstream.

field type req description
asset_param string ✓ FK into `params[].name` — the asset this branch reconstitutes when missing.
builder string ✓ Builder process or subworkflow name invoked inside the branch (e.g. `SAMTOOLS_FAIDX`, `BWA_INDEX`).
builder_outputs string[] ✓ Output channel names from the builder that are assigned to the asset (e.g. `fai`, `dict`).
evidence → Evidence ✓
guard string ✓ Verbatim guard expression, e.g. `!fasta_fai_in && step != "annotate"`.
fallback_for string | null Name of the `<asset>_in`-style param that, when present, supplies the asset directly and skips the rebuild. Null when no such pairing was detected.
guard_params string[] Param names referenced by the guard. Empty when the guard binds only non-param locals.

SampleSheet

SampleSheet

One sample-sheet-shaped pipeline input. Captures the row schema, path-vs-meta column classification, and discovery provenance so downstream Molds can map onto Galaxy `sample_sheet[:variant]` collections, CWL records-of-arrays, or other target shapes without re-reading the source.

field type req description
columns → SampleSheetColumn[] ✓ Row schema in declaration order. nf-schema's `samplesheetToList` emits columns in property order, not source-column order — preserve property order here so downstream translations match runtime channel item layout.
discovered_via string ✓ How the sample-sheet shape was detected. `nf-schema`: nextflow_schema.json entry has `schema:` keyword. `samplesheetToList`: workflow imports nf-schema's helper and calls it on the param. `splitCsv`: workflow uses `splitCsv(header: true)` on the param's path. `ad-hoc`: pipeline-specific CSV/TSV parsing inferred from script bodies.
param string ✓ Foreign key into `params[].name` — the parameter whose value points at the sample-sheet file.
format string | null Tabular format declared by the param's `mimetype` or inferred from `samplesheetToList` / `splitCsv` semantics. Null when undeclared.
header boolean | null Whether the file is expected to have a header row. Null when the discovery source doesn't pin it.
schema_path string | null Repo-relative path of the JSON Schema file describing rows (e.g. `assets/schema_input.json`), resolved from the param's nf-schema entry's `schema:` keyword. Null when discovered via `samplesheetToList` without a schema reference, or via `splitCsv(header: true)` ad-hoc parsing.

SampleSheetColumn

SampleSheetColumn

One column in a sample-sheet row schema. Type vocabulary is JSON Schema-style scalar; `kind` separates dataset-reference columns from per-row metadata. Validation hints (`pattern`, `enum`, `exists`, `mimetype`) are preserved verbatim — target Molds decide which of them survive translation (e.g. Galaxy's `sample_sheet` validator allowlist is regex/in_range/length).

field type req description
kind string ✓ `data`: path-typed column whose values are dataset references (becomes element/inner-collection slot in Galaxy `sample_sheet`). `meta`: scalar metadata column (becomes a `column_definitions` entry, populated per row in `columns[]`). Includes the nf-schema `meta:` annotation when present; in its absence inferred from `format` (`file-path`/`directory-path`/`path` → `data`, anything else → `meta`).
name string ✓
required boolean ✓
type string ✓ JSON Schema-style scalar type. Path columns are typed `string` with a `format: file-path|directory-path|path` qualifier — see `format` and `kind`.
default any Default value verbatim from the sample-sheet schema. May be null.
description string | null
enum any[] Allowed values when the column schema declares an `enum`. Empty array when absent.
exists boolean | null nf-schema `exists` flag for path columns. True when the parser must verify the file exists.
format string | null JSON Schema `format` keyword. nf-schema vocabulary: `file-path`, `directory-path`, `path`, `file-path-pattern`. Null for scalar metadata columns without a format hint.
mimetype string | null nf-schema `mimetype` keyword (e.g. `text/csv`, `application/gzip`). Useful for target datatypes.
pattern string | null JSON Schema regex `pattern` when present.

SnapshotChannel

SnapshotChannel

One output channel or flat snapshot list inside a `.snap` entry.

field type req description
files → SnapshotFile[] ✓ File md5 assertions parsed from `<path>:md5,<hex>` strings.
key string | null ✓ Channel key from the snapshot object, e.g. `0`, `1`, or `genome_gtf`; null for flat file-list snapshots.
values any[] ✓ Non-file snapshot values preserved verbatim for consumers that need versions, counts, or scalar tuple emissions.

SnapshotContent

SnapshotContent

One top-level snapshot entry from an nf-test `.snap` JSON sidecar.

field type req description
channels → SnapshotChannel[] ✓ Channel-keyed snapshot values. Pipeline-template flat file lists use `key: null`.
name string ✓ Top-level key in the `.snap` file, usually the `test("...")` name.

SnapshotFile

SnapshotFile

One file digest assertion from a `.snap` sidecar.

field type req description
basename string ✓ Final path segment, suitable for Galaxy test `file:` assertions when the snapshot stores only basenames.
md5 string ✓ MD5 digest from the sidecar.
path string ✓ Path portion before `:md5,` exactly as stored in the sidecar.
stub boolean ✓ True when the digest is the empty-file md5 emitted by many `-stub` tests.

SnapshotFixture

SnapshotFixture

Structured shape of an nf-test `snapshot(...).match()` assertion. nf-core templates pass a small set of values into snapshot() — succeeded-task count, version YAML (with Nextflow version stripped), stable file-name list, stable file-content list — and prune the comparison via `getAllFilesFromDir(..., ignoreFile: ..., ignore: [...])` helpers. This breakdown lets downstream consumers reconstruct equivalent assertions in target test frameworks without re-parsing Groovy.

field type req description
captures string[] ✓ Logical names of values passed to snapshot(), in order. Common set: `succeeded_task_count`, `versions_yml`, `stable_names`, `stable_paths`. Free-form strings rather than an enum because pipelines can pass other values.
helpers string[] ✓ nf-test helper functions invoked in the snapshot expression (e.g. `getAllFilesFromDir`, `removeNextflowVersion`). Used to detect whether a target framework can replicate the comparison.
ignore_files string[] ✓ Repo-relative paths passed as `ignoreFile:` to helpers (e.g. `tests/.nftignore`, `tests/.nftignore_files_entirely`). The files themselves are line-delimited globs.
ignore_globs string[] ✓ Inline `ignore: [...]` glob list passed to helpers (e.g. `['Prokka/**', '**/multiqc_busco.yaml']`). Distinct from ignore_files which references on-disk lists.
parsed_content → SnapshotContent[] Parsed entries from the `.nf.test.snap` JSON sidecar. Empty when the sidecar is absent, too large for compact summary output, or not valid JSON. Each entry preserves the snapshot name plus channel-keyed file md5 assertions and non-file scalar values.
snap_path string | null Repo-relative path of the corresponding `.nf.test.snap` file when present.

SourceRecord

SourceRecord

Provenance block. Mirrors gxy-sketches SketchSource field names so a Foundry summary is structurally consumable by anyone using that shape.

field type req description
ecosystem string ✓ Pipeline flavor. `nf-core` when `manifest.name` is `nf-core/<slug>` and the canonical layout is present; `nextflow` for ad-hoc DSL2 pipelines. Superset of gxy-sketches' enum (which adds `iwc`, `snakemake-workflows`, `wdl` for its other ingestors).
slug string ✓ Kebab-case identifier; `<owner>-<repo>` for nf-core, repo basename otherwise.
url string ✓ Git remote URL of the pipeline repository.
version string ✓ Tag, branch, or commit SHA the summary was derived from.
workflow string ✓ Pipeline name (typically the part after `nf-core/`).
license string | null SPDX license identifier when detectable from a LICENSE file.

Subworkflow

Subworkflow

One named `workflow <NAME> { ... }` block other than the primary workflow. Includes nf-core utility wrappers (PIPELINE_INITIALISATION, PIPELINE_COMPLETION, UTILS_NFCORE_PIPELINE) which exist to compose helper-function calls rather than to invoke processes.

field type req description
calls string[] ✓ Names of processes and nested subworkflows this subworkflow invokes. Free-function calls (e.g. `paramsHelp`, `completionEmail`) are NOT recorded here; they're implied by `kind = utility`.
kind string ✓ `pipeline` when the subworkflow invokes pipeline processes (data-flow contributor). `utility` when it composes helper functions only — common nf-core template subworkflows like PIPELINE_INITIALISATION (which calls `paramsHelp`, `samplesheetToList`, `UTILS_NFSCHEMA_PLUGIN`) and PIPELINE_COMPLETION (`completionEmail`, `completionSummary`). Utility subworkflows have empty `calls[]` for processes but may still expose channel outputs the primary workflow consumes.
name string ✓
path string ✓ Relative path from the pipeline root.
tests → NfTest[] ✓ Subworkflow-scoped nf-test files under the subworkflow's `tests/` directory. Empty when absent.
aliases string[] Alias names this subworkflow is imported as via `include { <NAME> as <ALIAS> }`. Calls retain the spelling used at each call site, while this field maps those aliases back to the canonical `name`. Empty when the subworkflow is never aliased.
inputs → ChannelIO[] Parsed entries of the subworkflow's `take:` block. `name` is the take name, `description` carries any adjacent `// ...` comment, `shape` carries the verbatim source line.
invocations → SubworkflowInvocation[] Call-sites where this subworkflow was invoked. Each entry records the calling workflow/subworkflow, the verbatim positional arguments, and a per-argument binding onto this subworkflow's `inputs[].name` (take names). Drives reference-asset attribution (`reference_assets[].used_by`) without re-parsing call sites. Empty when uncalled or when the call could not be positionally aligned.
outputs → ChannelIO[]

SubworkflowInvocation

SubworkflowInvocation

One call-site of a subworkflow, with positional arguments bound to its `take:` names.

field type req description
arguments string[] ✓ Verbatim positional argument expressions in source order (e.g. `params.fasta`, `PREPARE_GENOME.out.fai`).
bindings → InvocationBinding[] ✓ Per-argument binding onto the callee's `inputs[].name` (take names), in source order. Length matches `arguments[]` when the call positionally aligns to the callee's take block; shorter when only a prefix aligns.
caller string ✓ Name of the calling workflow or subworkflow.
caller_path string | null Repo-relative path of the file the caller is defined in. Null when not resolvable.

Summary

Summary

Top-level shape. Every Nextflow summary is exactly this object.

field type req description
params → Param[] ✓
processes → Process[] ✓
profiles → Profile[] ✓ Every profile declared in a `profiles { }` block, with the role(s) each one plays. Kinds are classified from the profile body (and its resolved `includeConfig` chain), not from the profile name: a name-shaped guess mislabels both directions — `replikation/What_the_Phage`'s `test` profile only flips a mode boolean, and `nf-core/sarek`'s `mutect` profile includes `conf/test_mutect2.config`. Across the 26-pipeline fixture corpus 103 of 371 profiles carry profile-level test-input evidence. This evidence feeds candidate discovery but does not replace test-case selection; mode profiles can also encode target-scope choices, while container and executor profiles describe the source execution environment.
reference_assets → ReferenceAsset[] ✓ Curated, target-agnostic view of reference-data inputs the source pipeline consumes (FASTAs, indexes, dictionaries, annotation tables, …). Each entry is a foreign-key projection onto `params[]` plus source-side metadata (asset kind, format hint, source expression, schema group, callers). Empty array when no reference assets are detected. Downstream Molds (e.g. nextflow-summary-to-galaxy-reference-data) classify these against Galaxy translation rungs; classification decisions are not encoded here.
reference_rebuilds → ReferenceRebuildRule[] ✓ Compute-if-missing rebuild branches detected in workflow/subworkflow bodies. Each entry binds an asset param to the guard and builder process that recomputes it when absent. Empty when no rebuild idioms are found. Source evidence only — Galaxy in-tool-rebuild decisions live downstream.
sample_sheets → SampleSheet[] ✓ Structured sample-sheet inputs. Each entry binds one `params[]` parameter to a row schema (column names, types, path-vs-meta classification, required flags, enums, patterns). Promoted from prose inside `params[].description` so downstream target translations (Galaxy `sample_sheet*` collections, CWL records-of-arrays) can choose collection variants without re-parsing the source pipeline. Empty array when no sample-sheet idiom is detected. Discovery sources: nf-schema `schema:` references, `samplesheetToList()` calls, and `splitCsv(header: true)` materializations.
source → SourceRecord ✓
subworkflows → Subworkflow[] ✓
test_candidates → TestCandidate[] ✓ The maximally useful statically recoverable whole-pipeline test cases. When pipeline-level nf-test exists, each literal `test(...)` block is a candidate (dynamically generated files retain one explicit aggregate candidate and a warning). Otherwise input-bearing profiles are candidates; otherwise pipeline defaults form the fallback candidate. Every candidate carries its own resolved profiles, parameter delta, inputs, assertions, scale/coverage classification, and selection disposition.
test_selection → TestSelection ✓ The deterministic selection result over `test_candidates[]`. A null selected id is an explicit scope-choice result, never an implicit fallback to the first filename or to a profile named `test`.
tools → Tool[] ✓
workflow → Workflow ✓
warnings string[] Cast-skill-emitted advisory messages (e.g. DSL1 detected, meta.yml/script disagreement, lossy assertion summarization). Empty array on a clean run.

TestCandidate

TestCandidate

One whole-pipeline execution candidate. The shape unifies nf-test cases, input-bearing profiles, and the no-profile defaults fallback so inputs and assertions never live in a separate singular record.

field type req description
assert_workflow_success boolean | null ✓ nf-test workflow-success assertion, or null for profile/default candidates.
disposition string ✓
effective_profiles string[] ✓ Resolved profile chain after nf-test.config, file/test directives, and an explicit caller override. `+name` directives append; other directives replace.
execution_mode string ✓
id string ✓ Stable candidate identifier: `<path>::<test-name>` for nf-test, `profile:<name>` for a profile, or `pipeline-defaults`. Duplicate names in one file receive a numeric suffix.
inputs → TestDataRef[] ✓ Input fixtures resolved for this candidate, including fetched samplesheet members when requested.
kind string ✓
name string ✓ Human-readable test name, profile name, or `pipeline defaults`.
outputs → ExpectedOutputRef[] ✓
params_delta object ✓ Candidate-specific parameter assignments after profile config and per-test overrides, excluding unchanged pipeline defaults.
path string | null ✓ Repo-relative `.nf.test`, profile-declaring config, or selected entrypoint path.
prose_assertions string[] ✓
rationale string ✓ Why the candidate exists and how its disposition was classified.
scope string ✓ Coverage/scale classification used by selection. Names are evidence: full-scale is deferred by default; minimal/tiny/stub is bootstrap-only.
snapshot → SnapshotFixture | null ✓

TestDataRef

TestDataRef

One test-fixture input. Field names mirror gxy-sketches' TestDataRef verbatim. The sketch-bundle constraint that `path` must live under `test_data/` is intentionally dropped; the Foundry summary describes fixtures as data, it does not bundle them.

field type req description
role string ✓ Logical role of the input (e.g. `samplesheet`, `reference_fasta`).
description string | null
filetype string | null File format, e.g. `fastq.gz`, `csv`.
path string | null Repo-relative path when the fixture lives in-tree, or local filesystem path when the CLI fetched the remote fixture into a caller-provided test-data directory.
sha1 string | null SHA-1 integrity hash when published alongside the fixture.
url string | null

TestSelection

TestSelection

Selection outcome over the complete candidate list.

field type req description
rationale string ✓
selected_candidate_id string | null ✓ Foreign key into `test_candidates[].id`; null exactly when status is needs-scope-choice.
status string ✓

Tool

Tool

One tool/dependency the pipeline runs. Mirrors gxy-sketches ToolSpec (name + version) and augments it with the resolved container/conda strings the bridge to author-galaxy-tool-wrapper needs.

field type req description
name string ✓ Canonical tool name (e.g. `fastp`, `samtools`). Used as the foreign key from `processes[].tool`.
version string ✓ Exact version when the conda spec pins one (`name=1.0`, `name==1.0`, or the space-separated `name 1.0`; a trailing `=<build>` is a conda build string and is not part of the version). The literal `unknown` when no exact pin was declared — either because the spec names no version at all (a legitimate authoring choice, e.g. `conda-forge::r-base`) or because it declares an inexact constraint, in which case `version_constraint` carries it. A spec the parser cannot read at all is reported in `warnings[]` rather than landing here as `unknown`.
bioconda string | null `bioconda::<name>=<version>` spec when a `conda` directive provides one — directly or by reference to a `${moduleDir}/environment.yml` whose `dependencies:` list contains a single bioconda entry. The latter is the modern nf-core convention.
biocontainer string | null BioContainers image reference when one was found. Includes both `quay.io/biocontainers/<name>:<version>--<build>` and the docker.io alias `biocontainers/<name>:<version>--<build>` (modern nf-core modules typically use the docker.io form in the docker branch and the depot.galaxyproject.org form in the singularity branch).
docker string | null Non-biocontainer Docker registry image string when present.
mulled_components → ToolSpec[] Constituent Bioconda packages for an opaque `mulled-v2-*` multi-package container when resolved from a cached BioContainers multi-package-containers TSV. Omitted when no matching cached row is available.
singularity string | null Singularity image reference when present (e.g. `https://depot.galaxyproject.org/singularity/...`).
version_constraint string | null Verbatim version constraint when the conda spec declares an inexact one (e.g. `>=1.17`, `>=1.0,<2.0`). Null when the spec pins exactly or names no version. Present so a range is preserved rather than flattened into the `unknown` sentinel; a consumer that needs an exact version should read a non-null constraint as "the pipeline constrains this tool, but not to a single version".
versions string[] Every distinct version of this tool declared anywhere in the pipeline, sorted, when processes disagree. nf-core modules pin independently, so one pipeline routinely runs several versions of a shared dependency (e.g. `samtools` at 1.17, 1.18, 1.19.2, 1.20). The sibling `version`, `biocontainer`, `docker`, `singularity`, and `wave` fields all come from whichever declaration was read last, so on a divergent tool they describe one arbitrary process rather than the pipeline as a whole — treat this array as the authoritative version set and read `processes[].conda` / `processes[].container` for what a given process actually runs. Declarations that name no version contribute nothing here, so a tool pinned by some processes and left unpinned by others lists only its real versions. Omitted when fewer than two distinct versions were declared.
wave string | null Seqera Wave / community-cr registry reference (e.g. `community.wave.seqera.io/library/<name>:<version>--<digest>` or `https://community-cr-prod.seqera.io/.../sha256/<digest>/data`). Wave-built containers are increasingly common in nf-core; kept distinct from `docker` because the resolution rules and provenance differ.

ToolSpec

ToolSpec

One constituent package in a decomposed multi-package container.

field type req description
bioconda string ✓ Exact Bioconda requirement spec, e.g. `bioconda::samtools=1.20`.
name string ✓
version string ✓

Workflow

Workflow

Primary workflow: channel construction plus the call-graph DAG. Real nf-core pipelines have multiple named `workflow` blocks (an anonymous entrypoint `workflow {}` in main.nf that wires PIPELINE_INITIALISATION → NFCORE_<NAME> → PIPELINE_COMPLETION; a named NFCORE_<NAME> wrapper; a substantive named workflow under workflows/<name>.nf). The summary collapses this into one primary `workflow` plus `subworkflows[]`. Selection rule: pick the workflow that invokes the most pipeline processes — typically the one under `workflows/<name>.nf`. The anonymous `workflow {}` glue and the NFCORE_<NAME> wrapper land in `subworkflows[]`.

field type req description
channels → Channel[] ✓
edges → Edge[] ✓
name string ✓ Name of the selected primary workflow (uppercase NF convention). The anonymous `workflow {}` entrypoint is never the primary; if it is the only workflow block, the pipeline is too small to summarize and the cast skill should emit a warning.
conditionals → Conditional[]

This page is auto-rendered from the JSON Schema authored in this repo and shipped on npm as part of @galaxy-foundry/summarize-nextflow (the producer co-locates its own schema). Each $def becomes a section below with a stable anchor ID — research notes and Mold bodies can deep-link individual shapes via summary-nextflow#Tool.

Source-of-truth chain:

  1. packages/summarize-nextflow/src/schema/summary-nextflow.schema.json — the canonical JSON, hand-edited as part of the Mold/cast loop (summarize-nextflow). Mold frontmatter cites it via summary-nextflow wiki-links; cast imports the summaryNextflowSchema runtime export and serializes it into cast bundles.
  2. packages/summarize-nextflow/scripts/sync-schema.mjs runs at prebuild, regenerating the typed summary-nextflow.schema.generated.ts const wrapper from the canonical JSON.
  3. Published as @galaxy-foundry/summarize-nextflow on npm. Site rendering imports the schema directly from this package via site/src/lib/schema-registry.ts; the published artifact also exports validateSummary() and ships the standalone summarize-nextflow bin (self-validates by default). The unified foundry CLI in @galaxy-foundry/gxwf-foundry exposes the same gate as foundry validate-summary-nextflow for downstream cast skills.

At runtime in cast skills: validation should happen through the CLI command:

foundry validate-summary-nextflow summary.json

The same schema is copied verbatim into references/schemas/summary-nextflow.schema.json per the casting policy in content/meta/casting.md. The package additionally exports validateSummary (AJV gate) for TypeScript consumers, but generated skills should prefer command-shaped validation so failures are easy to reproduce outside the agent runtime.

Contrast with tests-format, which is vendored from an external npm package (@galaxy-tool-util/schema); this schema is authored here and shipped to npm — the direction of the source-of-truth chain is reversed.

Why per-source

Paper, Nextflow, and CWL are different enough that forcing a shared cross-source summary shape would either lose detail or bloat all three (content/meta/harness-pipelines.md §“Mold-inventory parity”). Each summarize-<source> Mold emits its own schema; downstream source-target Molds such as nextflow-summary-to-galaxy-interface, nextflow-summary-to-galaxy-data-flow, nextflow-summary-to-cwl-interface, and nextflow-summary-to-cwl-data-flow consume this summary without pretending every source has one shared shape.

Field-name parity with gxy-sketches

Three sub-shapes mirror gxy-sketches verbatim — see gxy-sketches-alignment for the rationale:

  • SourceRecord — mirrors SketchSource (ecosystem, workflow, url, version, license, slug).
  • Tool — extends ToolSpec (name, version) with the resolved container/conda strings the bridge to author-galaxy-tool-wrapper needs.
  • TestDataRef / ExpectedOutputRef — mirror gxy-sketches’ field names exactly. The sketch-bundle invariant that path must live under test_data/ is intentionally dropped; the Foundry summary describes fixtures as data, it does not bundle them.

Cast-time role

Per content/meta/casting.md’s per-kind dispatch, this schema is referenced by summarize-nextflow via output_artifacts[].schema and copied verbatim into the cast bundle’s references/schemas/. The cast skill validates its emitted JSON with validate-summary-nextflow before returning; failure is loud — downstream Molds bind to this shape and would produce worse errors later.

What is intentionally not modeled

  • Structured channel typing. processes[].inputs[].shape is a string ("tuple(meta, [path,path])"), not a structured type. NF channel typing is a research project; a string is enough for downstream Molds to reason about and an LLM to emit.
  • Operator-chain semantics. Edge.via records the literal operator chain (["map", "join", "groupTuple"]). Reconciling what the chain does to channel shapes is left to the LLM step that fills Edge.notes when confidence is low.
  • Multi-tool processes outside decomposed mulled-v2 containers. A process can run multiple tools (a shell pipeline of two binaries). Process.tool is nullable; multi-tool processes set it null and surface tool details in script_excerpt and container. A tools[] foreign-key array on Process would be cleaner; deferred until downstream use forces it.

Revision 13 — 2026-09-15

Whole-pipeline test evidence is now one candidate model rather than two competing top-level representations.

  • test_candidates[] replaces test_fixtures and top-level nf_tests[]. Every candidate owns its resolved profile chain, parameter delta, inputs, outputs, execution mode, scope, disposition, rationale, and assertions.
  • test_selection makes ambiguity data. It either foreign-keys one selected candidate or records needs-scope-choice; consumers no longer infer a selection from array order or a profile named test.
  • Enumeration follows evidence fidelity. Literal whole-pipeline nf-test cases win, body-classified input profiles are the fallback, and pipeline defaults are the last candidate. Dynamically generated nf-test files retain a warned aggregate candidate instead of disappearing or requiring an nf-test runtime.
  • No silent CLI profile. --profile is optional and acts as a configuration override. Test description strings are not parsed for -profile text.
  • NfTest.execution_mode added. Module and subworkflow test evidence now records real/stub/mixed/unknown execution consistently with candidates.

Revision 12 — 2026-09-15

profiles[] became a classified list instead of bare names. Resolves galaxyproject/foundry#67 (a).

  • Profile shape added; profiles[] items are now objects, not strings. Each entry carries name, kinds[], source_path, includes[], and signals[]. This is a breaking change to the field; the two committed cast run artifacts were regenerated.

  • kinds is an array, not a scalar enum. 25 of the 371 corpus profiles select a container engine and an executor in one block — seven of biocorecrg/MOP2’s eight do, as does CRG-CNAG/CalliNGS-NF’s cluster, which sets process.container, singularity.enabled, and a process { } block carrying executor = 'crg'. A single kind would have had to lie about one of them.

  • test, container, executor, and dev accumulate; mode and resources do not. The latter two are assigned only when none of the former matched, because params-assignment and process-directive evidence is much weaker — nf-core’s docker profile assigns params.use_gpu inside a ternary, and most test profiles set resource caps. signals[] keeps the suppressed evidence visible.

  • Classification reads the profile body, never the name. The startsWith("test") heuristic consumers were forced into fails both directions: replikation/What_the_Phage’s test profile only flips fasta = true (now mode), while nf-core/sarek’s mutect includes conf/test_mutect2.config and nf-core/references’s cloud assigns params.input (both now carry test evidence). Corpus split over 371 profiles: 103 test, 195 container, 25 container+executor, 24 dev (7 of them alongside executor), 12 mode, 6 executor, 6 resources, 0 unknown.

  • test_full is not its own kind. It is kinds: ["test"] like any other test profile; includes carries conf/test_full.config, which is what a consumer separating cheap from release-scale candidates actually needs. 15 of 16 nf-core fixtures declare it.

  • NfTest.profiles[] stays string[]. Those are references to profile names, not declarations.

  • test is profile evidence, not the final selection. Pipeline-level nf_tests[] entries remain the higher-fidelity candidate unit because they preserve per-test parameter overrides. An input-bearing profile is the fallback when no such test case exists; its kind alone says nothing about runtime cost or representative coverage. Mode profiles may also carry target-scope choices, so they must not be treated like disposable executor or container settings.

  • test detection is scoped to nf-core input conventions. All 103 corpus test profiles are in the 16 nf-core fixtures; none are in the 10 ad-hoc ones, whose inputs are named reads / genome / variants. Absence of a test kind means ‘not detected’, not ‘none declared’, and the kinds description says so.

The enumeration fix this depended on shipped in the same change — see the summarize-nextflow rev 16 entry. A kind on a wrong name list would have been worse than no kind at all. One pipeline is still not fully enumerated: ncbi/egapx declares a profile in ui/assets/config/user/, outside the detected pipeline root, and reports none.

Revision 11 — 2026-09-08

Subworkflow topology fidelity from the pinned egapx fixture. Resolves galaxyproject/foundry#485.

  • Subworkflow.aliases: string[] added. Repeated include { workflow as alias } imports now map each call-site spelling back to the canonical subworkflow without collapsing the distinct calls.
  • Resolver structure extraction now ignores line and block comments. A commented future implementation with the same workflow name can no longer overwrite the live definition or leak its take:, emit:, and calls into the summary.

Revision 2 — 2026-05-01

First cast against nf-core/demo @ 1.1.0 exposed gaps in the v1 shape. Changes:

  • Tool.biocontainer description widened to accept the docker.io alias biocontainers/<name>:<version>--<build> alongside quay.io/biocontainers/.... Modern nf-core modules publish the docker.io form in the docker branch and the depot.galaxyproject.org/singularity/... form in the singularity branch.
  • Tool.wave field added for Seqera Wave / community-cr registry images (community.wave.seqera.io/..., https://community-cr-prod.seqera.io/...). Kept distinct from docker because resolution rules and provenance differ.
  • Process.container and Process.conda re-described as verbatim directive text (not “resolved”). Modern container directives are ternary expressions over workflow.containerEngine; modern conda directives are file references to ${moduleDir}/environment.yml. The schema now records the directive faithfully and pushes resolution into tools[].
  • ChannelIO.topic field added for Nextflow 24+ channel topics. nf-core templates emit a per-process topic: versions triple to a global topic for version aggregation; the v1 shape had no place to record this.
  • Subworkflow.kind enum added (pipeline | utility). nf-core template subworkflows like PIPELINE_INITIALISATION compose free-function calls (paramsHelp, completionEmail) without invoking processes; their calls[] is empty by design. The kind field disambiguates real data-flow contributors from utility wrappers.
  • Workflow.name description names the selection rule for pipelines with multiple named workflow blocks (anonymous workflow {} + NFCORE_ + a substantive named workflow): pick the one with the most process invocations — typically workflows/<name>.nf — and route the rest into subworkflows[].

What was not changed despite biting:

  • nf-test snapshot assertions (snapshot(...).match() with helpers) remain summarized to prose strings in ExpectedOutputRef.assertions[]. A structured “snapshot fixture” shape would help but is deferred until rev 3 when the testing note has paragraphs to inform the design.
  • Free-function calls in workflow bodies (paramsSummaryMap, softwareVersionsToYAML) remain folded into channel description text. No first-class representation; their effects are channel sources, the names are nf-core idiom not pipeline-specific signal.

Revision 3 — 2026-05-01

Second cast against nf-core/bacass @ 2.5.0 (33 processes, 9 nf-test files, 11 test profiles) exposed two structural-coverage gaps the second pipeline made universal. Changes:

  • Process.aliases: string[] added. Real pipelines re-import a single module multiple times under different aliases via include { X as Y } — bacass has six such patterns (CAT_FASTQ→{SHORT,LONG}, MINIMAP2_ALIGN→{CONSENSUS,POLISH} (3 aliases of one process), KRAKEN2_KRAKEN2→{KRAKEN2,KRAKEN2_LONG}, QUAST→QUAST_BYREFSEQID, plus FASTQC→{RAW,TRIM} inside FASTQ_TRIM_FASTP_FASTQC). Workflow edges[].from/.to reference alias names; canonical names didn’t appear in edges at all. The new field captures the alias→canonical mapping so downstream skills (especially author-galaxy-tool-wrapper, which needs to know “MINIMAP2_CONSENSUS shares MINIMAP2_ALIGN’s container/conda but is invoked with different runtime args”) can resolve references.
  • Summary.nf_tests: NfTest[] added. bacass has 9 tests/*.nf.test files, one per test profile. The previous schema’s test_fixtures is singular (one selected profile’s data shape); the rest of the test surface was invisible. The new array enumerates every .nf.test with structured fields: profile, params overrides, assert_workflow_success, prose_assertions, and a structured snapshot: SnapshotFixture | null capturing the snapshot(...).match() semantics.
  • SnapshotFixture shape added. nf-core templates use a near-uniform snapshot pattern: pass succeeded-task-count + version-yaml + stable-name list + stable-path list into snapshot(), pruning via ignoreFile: and ignore: globs. The new shape records captures, helpers, ignore_files, ignore_globs, and the .snap path — enough for downstream test-plan molds (e.g. nextflow-test-to-galaxy-test-plan) to reconstruct equivalent assertion intent in target frameworks without re-parsing Groovy.

What was not changed despite biting:

  • TestFixtures stayed singular. Multiple test profiles surfaced via nf_tests[] rather than promoting test_fixtures to an array — this preserves backward compatibility and keeps the “data shape of the selected profile” abstraction.
  • Mulled-v2 multi-package containers, multiMap/.branch/.cross fan-out, conditional channel construction, .mix-then-reassign — all still had only one bite each (bacass), so deferred per the “grow from contact” rule.

Revision 4 — 2026-05-05

Snapshot-sidecar parsing landed for module and subworkflow tests whose interesting assertions live in sibling .nf.test.snap JSON files. Changes:

  • SnapshotFixture.parsed_content: SnapshotContent[] added. Each parsed sidecar entry preserves the snapshot name plus channel-keyed SnapshotChannel values.
  • SnapshotFile added. <path>:md5,<hex> strings become file digest assertions with path, basename, md5, and a stub flag for empty-file md5s.
  • Non-file values preserved. Version tuples, counts, and other scalar snapshot values remain in SnapshotChannel.values so downstream test-plan Molds do not re-read .snap files.

Revision 10 — 2026-07-20

Conda-spec parsing fidelity: a partially-resolved directive is now either fully read or flagged. Resolves galaxyproject/foundry#356.

  • Tool.version_constraint added (string|null, optional — matching its sibling nullable fields, so cast artifacts emitted before this revision still validate). Verbatim inexact version constraint (>=1.17, >=1.0,<2.0) when the spec declares one; null when the spec pins exactly or names no version. Previously such a constraint was dropped and the tool collapsed to version: "unknown" alongside genuinely-unpinned tools.
  • Tool.version now reads == pins. The old pattern excluded = from the version group, so coreutils==9.4 fell through to "unknown". Corpus effect: findutils==4.6.0 in nf-core/createtaxdb and nf-core/proteinfamilies now records 4.6.0.
  • Tool.versions[] no longer contains the unknown sentinel. A tool pinned by some processes and left unpinned by others listed the sentinel as though it were a version. Corpus effect: nine tools across four pipelines cleaned; coreutils in nf-core/createtaxdb and nf-core/proteinfamilies now reads ["9.4", "9.5"], a real divergence the sentinel had masked.

Resolver-side changes with no schema surface: the pre-environment.yml ternary directive form conda (params.enable_conda ? "<spec>" : null) is now captured (previously invisible — it produced neither tools nor a warning); a conda spec that resolves in part now warns per unread spec rather than silently dropping it; and the null processes[].tool FK on genuinely multi-package processes is now a tested, documented decision rather than an untested silence.

Revision 9 — 2026-07-20

Reference-data extraction fidelity and tool-registry completeness, from a nf-core/eager cast. Resolves galaxyproject/foundry#349.

  • Tool.versions[] added (string[], optional). nf-core modules pin independently, so one pipeline routinely declares several versions of a shared dependency (nf-core/eager declares four samtools and six htslib versions). tools[].name is the processes[].tool foreign key, so the registry keeps one entry per name; the scalar version, biocontainer, docker, singularity, and wave fields all reflect whichever declaration was read last and therefore describe one arbitrary process on a divergent tool. versions[] is the authoritative version set; omitted when every declaration agrees. Purely additive — no existing field changed value.
  • ReferenceAsset.asset_kind gained reference_sheet for params naming a multi-reference sheet (nf-core/eager’s fasta_sheet), previously kinded other.

Resolver-side changes with no schema surface: legacy literal-string conda directives now populate tools[]; execution and registry params (input, outdir, multiqc_config, igenomes_base) no longer appear in reference_assets[]; used_by attributes params consumed via channel construction; and negative-guard rebuild detection covers the fused ch_x = BUILDER(args).chan assignment form.

Revision 7 — 2026-05-06

Top-level Param entries gained the nf-schema metadata previously only available on sample-sheet columns. Resolves galaxyproject/foundry#186.

  • Param.format added (string|null). nf-schema format keyword: file-path, directory-path, path, file-path-pattern. Disambiguates path-typed params from plain strings without re-reading nextflow_schema.json.
  • Param.hidden added (boolean|null). nf-schema hidden keyword. CLI-plumbing params (validate_params, pipelines_testdata_base_path, version, …) now drop out of user-facing target interfaces structurally.
  • Param.mimetype added (string|null). nf-schema mimetype keyword. Seeds Galaxy format on path params when present.
  • Param.schema_group added (string|null). The parent $defs section’s title (e.g. Input/output options, Reference genome options). Preserves nf-schema sectioning for UI grouping.
  • Param.fa_icon added (string|null). The parent section’s Font Awesome icon hint.

Downstream Molds — nextflow-summary-to-galaxy-interface, nextflow-summary-to-cwl-interface, nextflow-summary-to-galaxy-data-flow — can now consume these structurally instead of re-parsing the source schema. Mapping table in nextflow-params-to-galaxy-inputs.

Revision 6 — 2026-05-05

Sample-sheet schemas became first-class structured inputs. Resolves the open question raised in nextflow-workflow-io-semantics §“Open questions” and tracked in galaxyproject/foundry#177.

  • Summary.sample_sheets: SampleSheet[] added (required; empty array when none). Promotes sample-sheet shape out of params[].description prose so downstream target Molds can pick collection variants without re-parsing the source pipeline.
  • SampleSheet shape added. Binds one params[] parameter (param) to a row schema (columns) plus discovery provenance (discovered_via: nf-schema | samplesheetToList | splitCsv | ad-hoc), optional schema_path, format, and header.
  • SampleSheetColumn shape added. Captures name, JSON Schema-style scalar type, kind (data for path-typed dataset references vs meta for per-row metadata), format, required, default, enum, pattern, exists, mimetype, description. Validation hints stay verbatim — target Molds decide which survive translation (e.g. Galaxy’s sample_sheet validator allowlist is regex/in_range/length only; richer nf-schema validation downgrades to prose with confidence note).

Why now: nf-core’s samplesheetToList(params.input, "assets/schema_input.json") idiom maps almost 1:1 onto Galaxy’s sample_sheet[:paired|:paired_or_unpaired|:record] collection types (column_definitions, typed columns including element_identifier cross-row refs, restrictions[] for enums, regex validators). Without structured columns the interface Mold cannot pick sample_sheet:paired vs list:paired vs flat-file principally, and per-row meta fields silently fall to parallel parameter inputs. See galaxy-sample-sheet-collections for the target-side mapping table consumed by nextflow-summary-to-galaxy-interface and nextflow-summary-to-galaxy-data-flow.

What was not changed: Param.type still records the param’s own type (string/path) — the sample-sheet relationship is expressed by sample_sheets[].param referencing params[].name, not by mutating the param entry.

Revision 5 — 2026-05-05

Mulled-v2 multi-package container decomposition now has a narrow optional shape. Changes:

  • Tool.mulled_components: ToolSpec[] added. When summarize-nextflow is given a cached BioContainers multi-package-containers TSV, opaque mulled-v2-* container IDs can be decomposed into constituent Bioconda package specs.
  • ToolSpec added. Constituent packages record name, version, and exact bioconda requirement text.

Incoming References (21)