Skip to content

ai4curation/ai-gene-review / .claude/skills/module-curation

module-curation

Verified

Curate, review, or repair ai-gene-review ModuleReview YAML documents under modules/, including pathway/module boundary setting, parts and variant modeling, annoton-level molecular functions, representative UniProt/PTN grounding, module deep-research provenance, validation, rendering, and project batch updates.

Module Curation

Use this skill for modules/*.yaml work and for project pages that describe module/pathway curation. The goal is a reusable, defensible biological module, not a flat list of annotations or a species-specific note disguised as a module.

Workflow

  1. Read the local context before editing:
    • modules/README.md
    • the target modules/<module>.yaml
    • any adjacent <module>-deep-research-*.md
    • related projects/** batch page(s), if the task is pathway-batch work
    • relevant gene reviews and UniProt/GOA files for concrete members
  2. Decide the module boundary before filling YAML:
    • What biological process, complex, reaction chain, or reusable motif is the module?
    • Which genes/proteins are core members, and which are activation context, substrate supply, regulation, upstream/downstream biology, or separate modules?
    • Is this a concrete species/pathway instance or an abstract reusable motif?
    • Does the boundary have at least two substantive parts/roles/steps? If it collapses to one gene, one enzyme, or one reaction, do not create or retain a standalone ModuleReview; record it as pathway/gene curation and fold it into a broader module later.
  3. Model the structure:
    • Put module-level process/complex/context terms on module.concepts and module.context.
    • Put molecular functions on leaf annotons[].function, not on a biological-process module just because the member protein has that MF.
    • Use parts for required steps or roles, variant_sets for alternatives, and connections for ordered/causal links.
  4. Ground claims with checkable provenance:
    • Use exact UniProt, GO, Rhea, ChEBI, PANTHER PTHR, and PAINT PTN ids.
    • Never guess PTNs or identifiers. Resolve PTNs from local PANTHER/PAINT data or GOA WITH/FROM evidence.
    • Use representative members to orient family-level claims without turning a reusable module into a species-specific member list.
  5. Validate and render before committing:
    • uv run linkml-validate -s src/ai_gene_review/schema/gene_review.yaml -C ModuleReview modules/<module>.yaml
    • uv run python -m ai_gene_review.validation.module_validator modules/<module>.yaml
    • just render-module modules/<module>.yaml
    • render any touched project page with uv run ai-gene-review render-projects ...
    • run git diff --check

Reference Files

Load only the reference needed for the decision at hand:

  • references/modeling.md — module-vs-annoton placement, parts, variants, locations, exemplars, PTNs, and common anti-patterns.
  • references/research-and-grounding.md — module deep research, species/pathway satisfiability checks, gene review integration, and provenance rules.
  • references/validation-and-pr.md — validation, rendering, derived QC, cache noise, and PR checklist.

Curation Rules

  • Do not create or retain one-part modules. A standalone module needs at least two substantive parts/roles/steps at the modeled boundary; otherwise keep the work as pathway/gene curation or fold it into a broader multi-part module.
  • Avoid redundant parent/child context such as both cytoplasm and cytosol unless the distinction is intentional and explained.
  • Do not use broad parent processes as the module core when a specific process or reaction step is available. Keep broad terms as prose/context if needed.
  • Keep species-specific notes in evidence, notes, project pages, or concrete instantiations; reusable modules should stay species-neutral unless the scope is deliberately concrete.
  • Treat deep research as retrieval support, not authority. Every structural claim must still be grounded in identifiers, gene reviews, GOA/UniProt, GO-CAM, literature, or explicit curator judgment.
  • Record every ontology request in a structured field, never only in prose. A missing term (no specific process, no fitting activity) goes in the module's top-level proposed_new_terms, with proposed_name, proposed_definition, justification, a verified proposed_parent and supported_by quotes. A fault in an existing term goes in proposed_ontology_changes. Notes may point to the entry but must not be the only record. export_proposed_terms.py collects the top-level proposed_new_terms of modules and gene reviews; it does not read knowledge_gaps[].proposed_terms, which belong to a gap with a genuine biological unknown. Put a pure ontology request at the top level.
add-product-e2e-evalpaperclipai/paperclipAdd or extend a Paperclip full-stack runner E2E workflow, fixture, matcher, or report evidence path for local or Daytona execution.Stars: 115kUpdated Claude CodeCodexVerifiedInstall
add-runner-evalpaperclipai/paperclipAdd or extend a Paperclip Runner protocol evaluation definition, roster, assertion, or report fixture with provenance and narrow validation.Stars: 115kUpdated Claude CodeCodexVerifiedInstall
diagnose-why-work-stoppedpaperclipai/paperclipCompatibility entry point for failed, stalled, looping, or over-recovered Paperclip work. Use fix-failed-paperclip-run for the full diagnosis, regression, repair, real verification, and green-PR workflow; honor an explicitly requested diagnosis-only scope.Stars: 115kUpdated Claude CodeCodexVerifiedInstall
fix-failed-paperclip-runpaperclipai/paperclipDiagnose and fix failed, stuck, slow, or incorrectly blocked Paperclip tasks and runs. Use with a task/run URL or failure report to trace the root cause, prove a regression red then green, check second-order effects, verify the real workflow, and deliver a PR with passing checks.Stars: 115kUpdated Claude CodeCodexVerifiedInstall