feat(#1-3): meta-map contract + INPUT_CHECK + repo skeleton

[hyphaltip]

---

Summary

Phases 0–1 of the DSL2 modularization plan (REFACTORING_PLAN.md): replace the fragile positional 10-tuple data contract with the standard meta-map, introduce INPUT_CHECK to
deduplicate samplesheet parsing, and establish the modules/local/ / subworkflows/local/ directory skeleton.

Closes #1 (meta-map contract), #2 (INPUT_CHECK subworkflow), #3 (repo skeleton)

---

Issue #1 — atomic meta-map conversion (799897a)

Every process was wired with a positional 10-tuple that required touching every call site to add a field. Replaced with the standard DSL2 contract throughout:

Before: tuple(out, asmid, species, strain, locustag, busco, header_length, transl_table, gz, taxonid)
After: tuple val(meta), val/path(genome)
meta = [id, asmid, species, strain, locustag, busco, transl_table, taxonid]

• All 20 inline processes converted — shim pattern: input:/output: rewritten, alias block (def out = meta.id; def asmid = meta.asmid; …) at the top of each script: so existing
  shell interpolations are untouched.
• All ~40 workflow channel ops updated: filters, maps, groupTuple, combine, join.
• params.header_length (default 24) replaces the per-sample tuple field.
• Key non-obvious fix: clean_genome_ch uses it[0] index access instead of fixed-arity destructuring because .combine(ifEmpty([])) produces a variable-length tuple (2 elements when
  the sentinel is an empty list, 3+ when populated). A fixed { meta, _gz, _sentinel -> } closure fails on the 2-element case.

Issue #2 — INPUT_CHECK shared subworkflow (53deaef)

Extracted the samplesheet ingestion logic (duplicated between funannotate.nf and earlgrey_mask.nf) into subworkflows/local/input_check.nf. Emits two channels:

• samples — val(meta): taxon / asmid / suppress / n_test filters applied, no genome-existence check (used for postpredict, earlgrey select path)
• genomes — tuple(val(meta), path(gz)): additionally filtered to samples whose raw genome file exists (the primary jobs channel in funannotate)

earlgrey_mask.nf's skip_select_reps=true path now calls INPUT_CHECK and gains taxon/asmid/suppress/n_test filtering it previously lacked entirely.

Issue #3 — repo skeleton (53deaef)
Directory layout established:
modules/local/ ← one process per file (convention)
subworkflows/local/ ← shared subworkflows
workflows/ ← future workflow entrypoints

• modules/asm_stats.nf → modules/local/asm_stats.nf (git mv)
• Three annotation processes extracted from the funannotate.nf monolith into individual module files: antismash_run.nf, interproscan_run.nf, signalp_run.nf
• Stale modules/annotation_tools.nf deleted (old 10-tuple versions, superseded by the above)

---

Test plan

• nextflow run . -profile test -stub-run — 15 completed, 0 failed (full pipeline graph: GENOME_CLEAN_BATCH → MASKREPEAT×2 → FUNANNOTATE_PREDICT×2 →
  ANTISMASH/INTERPROSCAN/SIGNALP×2 → FUNANNOTATE_ANNOTATE×2)
• nextflow run . -profile test -stub-run --run_sra_fetch true — RNA-seq subgraph validation (dual-stub)
• Real-data HPCC run — confirms grouping picks the right representative and reads join to the right strain

🤖 Generated with Claude Code

https://claude.ai/code/session_01AbWwmm9v6dJn13sgAJn3Sa

• Fix CI-breaking params.genome_dir; repivot provisioning to ucr_hpcc
• Replace modularization status docs with authoritative plan + issue tracking
• meta-map groundwork: SampleUtils.makeMeta, params.header_length, recipe
• feat(Adopt meta-map data contract (Phase 0, BLOCKER) #3): atomic meta-map conversion — replace 10-tuple with tuple(val(meta), val/path(genome))
• feat(EPIC: Modularize nf_funannotate1 into DSL2 modules + subworkflows #2,Adopt meta-map data contract (Phase 0, BLOCKER) #3): INPUT_CHECK subworkflow + repo skeleton (modules/local/, subworkflows/local/)