summarize-nextflow: skip-flag guards produce false-positive reference_rebuilds and leak into reference_assets

[jmchilton]

Filed by Claude (AI assistant) on behalf of @jmchilton, from a review of the #349 fixes.

detectReferenceRebuilds in the summarize-nextflow resolver emits rebuild rules for boolean skip-flag guards that have nothing to do with reference data. The bogus params then leak into reference_assets[], so a downstream reference-data Mold sees skip_fastqc presented as a reference asset built by FASTQC.

Pre-existing on main; unaffected by the #349 work (identical output before and after).

Repro

foundry summarize-nextflow workflow-fixtures/pipelines/nf-core__bacass --no-with-nextflow

reference_rebuilds:
  asset_param=baktadb      guard=!ch_path_baktadb && val_baktadb_download  builder=BAKTA_BAKTADBDOWNLOAD   # legitimate
  asset_param=skip_fastqc  guard=!val_skip_fastqc                          builder=FASTQC_RAW              # bogus
  asset_param=skip_fastp   guard=!val_skip_fastp                           builder=FASTP                   # bogus
  asset_param=skip_fastqc  guard=!val_skip_fastqc                          builder=FASTQC_TRIM             # bogus

reference_assets: ['baktadb', 'skip_fastp', 'skip_fastqc']

3 of the 4 rules are false positives, and 2 of the 3 assets.

Root cause

The detector already tries to filter this shape:

if (/^(skip|no|disable|disabled)_/u.test(negated)) continue;

but it tests the negated identifier, which in bacass is val_skip_fastqc — a val_-prefixed take slot, not skip_fastqc. The prefix defeats the anchored regex, the rule is emitted, and resolveAssetParam then resolves it back to the real skip_fastqc param.

Two independent gaps:

1. The skip-flag filter is defeated by wrapper prefixes. nf-core subworkflows conventionally rename take slots (val_, ch_, _in). The filter should strip known prefixes before matching, or better, test the resolved asset_param rather than the raw negated identifier.
2. reference_assets[] trusts reference_rebuilds[] unconditionally. In buildReferenceAssets, params reaching the inventory via rebuildParams bypass the isNonReferenceParam filter and the isPathTypedParam check entirely. A boolean param should never become a reference asset regardless of how it got nominated — the asset inventory should require the param be path-typed (or at minimum non-boolean) even on the rebuild path.

Fixing (2) alone would keep reference_assets[] clean while leaving the bogus rules in reference_rebuilds[]; both are worth doing.

Suggested acceptance criteria

• bacass emits exactly one rebuild rule (baktadb / BAKTA_BAKTADBDOWNLOAD) and one reference asset (baktadb).
• A boolean-typed param cannot appear in reference_assets[] via any path.
• The skip-flag filter survives val_ / ch_ / _in take-slot renaming.
• sarek (6 rules) and eager (5 rules) are unchanged — both are currently correct.