Personal Genomics

## Quick judgment A powerful, privacy-preserving local genomics pipeline that is highly capable for personal DNA exploration and agent-assisted summarization, but inherently high-stakes and easy to misuse if the AI (or user) treats the output as clinical diagnosis or medication guidance instead of as informational input. ### Strengths - **Very broad scope**: >1600 markers across pharmacogenomics, hereditary cancer, carrier status, autoimmune HLA, polygenic risk, traits, fitness, nutrition, longevity, and ancestry/haplogroups. - **Agent-ready outputs**: `agent_summary.json` is well-structured (critical alerts, pharmacogenomics alerts, APOE status, PRS, ancestry, lifestyle recommendations, data quality), enabling an AI to triage and explain results. - **Privacy-first**: All analysis is local; no network calls. This is a major advantage for genomic data. - **Clinical-adjacent features**: Pharmacogenomics with CPIC guideline integration, FDA warning flags, hereditary cancer panel, and a genetic-counselor export, plus physician-shareable PDF with clear disclaimers. ### Key risks & safety considerations - **Medical decision risk**: Findings include pharmacogenomic dosing implications and hereditary cancer variants. If the AI uses this skill to *recommend or change medications, screening schedules, or treatments*, that crosses into clinical practice and is unsafe. - **Interpretation complexity**: Polygenic risk scores, HLA associations, and probabilistic disease risks are easy to misinterpret as deterministic. Users may overestimate their risk or be falsely reassured by “normal” results. - **Consumer array limitations**: The documentation notes that arrays miss rare variants and do not rule out conditions (especially cancer syndromes). An AI must repeatedly emphasize this; “no variant found” does **not** mean “you are clear.” - **Psychological impact**: High-risk results (e.g., BRCA1/2, APOE ε4/ε4, HLA-B27) can cause distress or anxiety. The agent should be cautious, provide context, and encourage professional genetic counseling. - **Population and ancestry nuances**: Risk estimates and allele frequencies are population-specific. If the user’s ancestry is mixed or underrepresented, the AI must highlight that risk estimates may be less reliable. ### Recommended scenarios Use this skill when: - A user explicitly provides **raw DNA data** (23andMe/AncestryDNA/MyHeritage/FTDNA or VCF) and asks for detailed, privacy-preserving analysis. - The goal is **education and preparation** (e.g., “to discuss with my doctor/genetic counselor”), not to make independent medical or medication decisions. - The agent can: - Clearly present critical findings and their limitations. - Repeatedly remind the user that this is **not a diagnosis** and cannot replace clinical testing or professional advice. - Encourage follow-up with healthcare professionals, especially for pathogenic hereditary cancer variants, APOE ε4/ε4, multiple high-impact pharmacogenomic findings, or results causing significant distress. Avoid or heavily constrain this skill when: - The user is seeking **immediate treatment changes, dosing changes, or screening decisions** based solely on these results. - The user is in acute distress or appears unable to handle probabilistic or uncertain results without support. - Regulatory, clinical, or organizational policies require **certified clinical-grade tools or labs** for genetic testing and interpretation. Overall, this is a high-functionality, high-privacy genomics tool that is appropriate for careful, well-qualified agent use in an informational and supportive role, but it demands strict guardrails around medical advice, interpretation, and user emotional well-being.