Brief for module-curation agents (YeastPathways module project)

Repo: ai-gene-review. You are creating or revising modules/<slug>.yaml
documents that generalise YeastPathways (SGD YeastCyc) pathways into reusable
biological modules. This is NOT a rote translation of YeastCyc.

Read first

• CLAUDE.md (identifier discipline; the NEW-annotation and IBA rules apply to
  module claims too).
• .claude/skills/module-curation/SKILL.md and its references/*.md.
• modules/README.md.
• Your group(s) in projects/YEAST_PATHWAYS/crosswalk.yaml — the action
  (NEW / ALIGN / GENERALIZE), the chosen scope and the rationale. You may
  disagree with the rationale after reading the literature; if so, explain in
  your final answer and in the module notes, but keep the slug.
• The source pathway(s) in projects/YEAST_PATHWAYS/data/yeastpathways_summary.yaml
  (reactions, EC numbers, catalysts with UniProt ids, YeastCyc citations, comment).
• The GO-CAMs listed for the group under gocams/<id>/<id>-src.yaml
  (YeastPathways_* = SGD conversion; hex ids = PomBase S. pombe models).
  Use the gocam-curation skill conventions for reading them.
• For ALIGN/GENERALIZE: the existing module in full, plus 1-2 sibling modules of
  similar quality (e.g. modules/histidine_biosynthesis.yaml,
  modules/coenzyme_a_biosynthesis.yaml).

What to produce

1. Critical pathway assessment. For each YeastCyc reaction decide whether it
   belongs in the module, which enzyme really catalyses it (YeastCyc gene lists
   are incomplete or stale in places: e.g. missing ARO10, SOL3/SOL4, GND1/2, THI4,
   MEU1/MRI1/MDE1/UTR4/ADI1, GLG1/2, DUG1-3), and the compartment. Find the
   primary literature: use WebSearch / the article tools, then cache each cited
   paper with just fetch-pmid <PMID> so publications/PMID_<n>.md exists
   before you cite it. Only cite PMIDs whose cached file you have read.
2. Generalisation. Model at the scope in the crosswalk: species-neutral
   label/description, parts in pathway order, variant_sets for real
   alternatives (architecture fusions, route variants, lineage-specific steps),
   connections with PRECEDES (+ chaining_status/chaining_note when the
   CoA check would mis-fire). MF terms on leaf annotons only. Module-level
   concepts: the most specific GO BP term(s).
3. Exemplar grounding. Every leaf role gets representative_members (or a
   concrete GENE_PRODUCT participant) that include the S. cerevisiae S288C
   protein(s) (UniProtKB accession from the summary YAML or the gene's UniProt
   file) and, when the group lists a PomBase GO-CAM, the S. pombe ortholog(s)
   (accessions from the PomBase GO-CAM / UniProt). Keep existing non-fungal
   representatives. Use a FAMILY selector with a PANTHER family id only if you
   have verified it (grep -A1 "^id: PANTHER:PTHRxxxxx$" interpro/panther/panther.obo
   and membership); otherwise omit term and keep preferred_term (see CLAUDE.md
   "PANTHER ids"). Never guess PTN ids.
4. GO-CAM grounding. Add gocam_associations on the module node for each
   listed model (with title), and on leaf annotons where a specific
   activity id clearly corresponds (format gomodel:<model>/<activity>; for
   YeastPathways models the activity ids look like gomodel:HISTCYCLOHYD-RXN
   -- copy them exactly from the src file). Where the GO-CAM disagrees with your
   module (wrong MF, missing step, extra gene), say so in notes.
5. Evidence: top-level evidence must cite YeastPathways:<frame id> for each
   source frame (url https://pathway.yeastgenome.org/YEAST/NEW-IMAGE?object=<id>),
   GO terms used as concepts, and the PMIDs.
6. Exemplar gene list. Append (or create) a section for each of your modules
   in projects/YEAST_PATHWAYS/module_members/<slug>.yaml:
   ```yaml
   module:
   scerevisiae: # every S. cerevisiae gene product used anywhere in the module
   • {symbol: HIS1, uniprot: P00498}
     spombe: # every S. pombe gene product used (only if a PomBase GO-CAM is linked)
   • {symbol: his1, uniprot: ...} # PomBase symbol as used for genes/SCHPO/
     panther_families: [PTHR...] # verified family ids used as role families
     ```
     This file drives which genes and families get reviewed, so include every
     exemplar gene product, including ones you added beyond YeastCyc.

Validate

uv run linkml-validate -s src/ai_gene_review/schema/gene_review.yaml -C ModuleReview modules/<slug>.yaml
uv run python -m ai_gene_review.validation.module_validator modules/<slug>.yaml
just render-module modules/<slug>.yaml

Errors must be fixed. Warnings about missing gene/family reviews are expected
(those reviews are being produced in parallel) -- do not create gene reviews.
Do not commit, push, or edit files outside modules/<your slugs>.yaml,
pages/modules/<your slugs>.html, projects/YEAST_PATHWAYS/module_members/,
and publications/ (via fetch-pmid). Revert incidental cache/ changes
(git checkout -- cache/).

Also create a history record per module:
just new-history --kind module --slug <slug> --event CREATE|EDIT --outcome changed --summary "..." --agent-tool claude-code --details "..."
and just validate-history <path>.

Final answer

Per module: action taken, boundary/generalisation decision, parts and
variant sets, YeastCyc corrections (missing/wrong enzymes, compartments),
GO-CAM disagreements, validation status, and counts of S. cerevisiae /
S. pombe exemplars.

Note: the scratchpad is shared by parallel agents; use a uniquely named
subfolder for any helper scripts.