[LifeSci Skill] Bound target-evidence retrieval

[phorward-deployed]

Summary

• add a bounded target-evidence workflow shared by the Entrez and ClinicalTrials.gov skills
• define consistent evidence slots for biology, positive and negative programs, human safety, normal-tissue context, and preclinical evidence
• batch selected PubMed records, reuse registry intervention aliases, and stop once the requested evidence categories are covered
• keep the change instruction-only; the existing API wrappers are unchanged

Evaluation

Same-prompt local testing showed materially lower latency and tool fan-out for a single-target evidence review. Multi-target testing also reduced tool calls and context growth, although end-to-end wall time remained variable. This is a draft so the retrieval limits and evidence coverage can be evaluated across additional targets before landing.

Tests

• quick_validate.py plugins/life-science-research/skills/ncbi-entrez-skill
• quick_validate.py plugins/life-science-research/skills/clinicaltrials-skill
• git diff --check