Produce a private, printable DNA analysis PDF for the person who is asking, using the genome.sh CLI on their local machine. Do the work, not a plan.

The deliverable is an A4 PDF report filled from their own DNA file (VCF, 23andMe, Ancestry, MyHeritage, or similar). It should match the genome.sh printable report: cover, methodology, five takeaways, domain dashboard, then heart, cancer, longevity, metabolism, immune, brain, physical traits, pharmacogenomics, recommendations, limits, and glossary. Optional extra panels: polygenic scores and trait grid when the file supports them.

1. Privacy

Keep the raw DNA file, intermediate TSV/JSON, filled HTML, and PDF on the user's machine. Never upload genomes to genome.sh, to https://api.genome.sh, or to any third party. Do not commit filled reports. Copy the template out of the repo before writing personal data:

```sh
cp -R docs/report-template ~/my-genome-report
cd ~/my-genome-report
```

Template and PDF helper: https://github.com/romainsimon/genome-sh/tree/main/docs/report-template

2. Analyse locally with genome

Install with `cargo install genome-sh` (or `conda install -c bioconda genome-sh`). The binary is `genome`.

```sh
genome db install standard
genome db status
genome query rs1799945 --format json
genome annotate INPUT.vcf.gz --format json
genome annotate INPUT.vcf.gz --filter clinical --format json
```

Use `standard` (or `full`) so ClinVar and gnomAD frequencies are available. Prefer `--format json` or `compact` for your own scripts. If the input is a consumer chip export, convert or parse it to genotypes first, then query the rsIDs that are actually present. State the assay type, SNP count, and reference build in the methodology section.

3. What the report must contain

Copy `report.html` and `generate-pdf.sh` from the template. Replace every `{{PLACEHOLDER}}`. Use status pills `favorable`, `watch`, `action`, or `neutral`. Keep the A4 print CSS.

Required narrative:
- Cover: name only if the user supplied it, date, sequencing source, tools (`genome` + ClinVar/gnomAD/PharmGKB).
- Methodology: chip vs whole-genome limits, missing major markers, sex chromosomes if known. A SNP chip is not a full BRCA screen.
- Five takeaways a non-specialist can act on or understand.
- Dashboard one-liners by domain.
- Domain findings with rsID, genotype, ClinVar significance, condition, and population frequency when returned.
- Pharmacogenomic card for drug-metabolism genes that were actually called.
- Recommendations split by strength, each tied to a finding.
- Limits and glossary.
- Medical disclaimer: informational, not a diagnosis, not a medical device.

4. Evidence rules

- Do not invent pathogenicity, frequencies, or gene-disease links. Use what `genome query` / `genome annotate` returns.
- If a famous marker is absent from the file (for example APOE ε4, a CYP star allele, or a founder BRCA variant), write "not called" instead of assuming wild type.
- Distinguish heterozygous, homozygous reference, and homozygous alternate.
- Chip files miss rare variants. Say so. Do not claim "no BRCA risk" from a few founder SNPs.
- AlphaGenome prediction is opt-in and sends sequence off-machine. Skip it unless the user explicitly asks.

5. Render the PDF

Fill `report.html`, then:

```sh
./generate-pdf.sh report.html report.pdf
```

Confirm no `{{PLACEHOLDER}}` remains. Open or list the PDF path. Tell the user where the private folder is and that it should stay off git.

Canonical documentation:
- https://genome.sh/docs/agent-guide.md
- https://genome.sh/llms.txt
- https://genome.sh/genome-catalog.json
- https://github.com/romainsimon/genome-sh/tree/main/docs/report-template
- https://crates.io/crates/genome-sh
- https://github.com/romainsimon/genome-sh