Skip to content

feat(skills): add public variant evidence routing - #3

Open
vladimir-mitev wants to merge 7 commits into
fmschulz:mainfrom
vladimir-mitev:feat/public-variant-evidence-routing
Open

vladimir-mitev wants to merge 7 commits into
fmschulz:mainfrom
vladimir-mitev:feat/public-variant-evidence-routing

Conversation

@vladimir-mitev

Copy link
Copy Markdown

Summary

Adds a focused bio-public-variant-evidence skill and routes eligible public
single-variant questions to it from the omics scientist agent.

The workflow uses Folklore Clinical Variant Interpretation MCP, published by
Helena Bioinformatics, for source-linked public variant evidence and
variant-linked literature. It preserves the live resolution outcome, stops on
ambiguity, and chains literature only after a variant is resolved. It also
routes broad annotation, arbitrary database lookup, VCF-wide analysis, and
unsupported variant classes to the repository's existing skills instead of
expanding this skill's scope.

The skill accepts one public GRCh38 germline nuclear SNV or simple indel. It
does not accept patient, phenotype, family, segregation, or private case data.
Results require qualified professional review and are not a diagnosis or
treatment recommendation.

Changes

  • add skills/bio-public-variant-evidence/SKILL.md with explicit inputs,
    outputs, outcome branching, composition rules, and quality gates;
  • register the skill in the omics scientist's mandatory usage, workflow tree,
    and recognition patterns;
  • add positive routing regressions for public variant evidence and
    source-linked publications, plus a negative VCF/VEP/ClinVar regression;
  • rebuild the generated catalog and routing baseline;
  • update the reader-facing skill index and repository counts.

Validation

  • python scripts/validate-skills.py: passed
  • python scripts/validate-supplementary-docs.py: passed
  • routing benchmark: 67/67, baseline 67/67, delta +0
  • strict MkDocs build: passed
  • generated catalog and routing baseline: deterministic
  • security scan with behavioral analysis: SAFE, no low-or-higher findings
  • live public tools/list: returned the four documented tools
  • live public rs80357914 smoke test: ambiguous, two candidates, no automatic selection

The full test suite reports 291 passed and 4 failed on macOS. The same four
failures reproduce on the unchanged main commit and are caused by existing
GNU realpath -m assumptions and macOS /var versus /private/var path
aliasing; none touches this skill or routing change.

Public references

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant