Skip to content
Back to skills

Trio Calling

ASecurity

"Plans DeepTrio trio and duo variant-calling workflows with child

  • 247 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 8, 2026
businesspythonbashdocker

Security analysis

A100/100

Pro scans all 5 files and shows the line behind each finding

Scanned September 8, 2026

npx -y skills add VectorSpaceLab/AREX-Skill --skill trio-calling --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Trio Calling?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Trio Calling
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/vectorspacelab-trio-calling/badge)](https://www.skillsdirectory.com/skills/vectorspacelab-trio-calling)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: trio-calling
description: "Plans DeepTrio trio and duo variant-calling workflows with child
  and parent inputs, per-sample VCF/gVCF outputs, and GLnexus merge handoff."
disable-model-invocation: true
metadata:
  disco-role: operating
license: BSD 3-Clause
---

# DeepTrio Trio Calling

Use this sub-skill when a user needs DeepTrio family calling for a child with one or two parents. It covers Docker/Bazel-style `run_deeptrio` command planning, duo behavior when one parent is absent, model-type choice, per-sample output planning, gVCF merge handoff, and family-specific troubleshooting.

## Route The Request

- Use this sub-skill for DeepTrio `run_deeptrio` workflows, trio/duo sample planning, child/parent output naming, gVCF planning, GLnexus handoff, and sex-chromosome parent-selection caveats.
- Route single-sample `run_deepvariant` germline calling to the `germline-calling` sub-skill.
- Route pangenome-aware DeepVariant workflows to the `pangenome-aware-calling` sub-skill.
- Route low-level `make_examples`, `call_variants`, `postprocess_variants`, sharded TFRecords, and extra-args internals to the `pipeline-stages` sub-skill.
- Route benchmarking reports, Mendelian-violation analysis, `hap.py`, RTG, and visualization to the `analysis-visualization` sub-skill.

## Fast Path

1. Identify family mode: trio has `child`, `parent1`, and `parent2`; duo has `child` and exactly one parent.
2. Choose `--model_type` from `WGS`, `WES`, `PACBIO`, or `ONT` according to the sequencing data and image/model availability.
3. Require an indexed FASTA and sorted, indexed, compatible BAMs for every supplied sample.
4. Require one VCF output for every supplied sample, and request a matching gVCF for every supplied sample when GLnexus or cohort merging is planned.
5. Use [`scripts/deeptrio_command_builder.py`](scripts/deeptrio_command_builder.py) to validate the plan and print a reviewable command without executing Docker, DeepTrio, or GLnexus.
6. Read [`references/workflows.md`](references/workflows.md), [`references/deeptrio-flags.md`](references/deeptrio-flags.md), and [`references/troubleshooting.md`](references/troubleshooting.md) for command patterns, flag contracts, and recovery steps.

## Bundled Helper

The helper is a command planner only:

```bash
python scripts/deeptrio_command_builder.py --help
```

It validates child/parent flag groups, makes duo mode explicit by omitting every parent2 flag, warns about WES regions and long-read candidate partitioning, and can emit a GLnexus handoff command when a complete gVCF set is present.

## Safety

Do not run Docker, GLnexus, network downloads, GPU jobs, or native DeepTrio tests without explicit user approval and real user-provided genomics inputs. Treat generated commands as plans until the user confirms the runtime environment, container mounts, and data paths.

Files in this skill

  • SKILL.md2.8 KB
  • references/deeptrio-flags.md6.5 KB
  • references/troubleshooting.md5.7 KB
  • references/workflows.md6.9 KB
  • scripts/deeptrio_command_builder.py12.9 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.