ACMG Variant Classification

---

name: acmg-variant-classification

description: Standard workflow for ACMG/AMP germline small-variant classification — collect evidence, assign criteria, detect conflicts, and produce a review-ready classification summary. Use when a user wants a structured ACMG/AMP-style interpretation workflow for a germline SNV/indel, including guided evidence intake, criteria assignment, conflict handling, and provisional classification.

---

# ACMG Variant Classification

Use this skill when a user wants a structured ACMG/AMP-style interpretation workflow for a germline SNV/indel.

Interaction mode

Default to a guided interview workflow.

When using this skill with a live user:

1. Ask for one block of information at a time

2. Wait for the user's answer before moving on

3. Do not request all evidence at once unless the user asks for a bulk template

4. Explicitly track what is known, unknown, and still needed

5. Treat phenotype, family history, segregation data, and parental genotypes as user-supplied inputs that may arrive incrementally

Recommended guided sequence:

1. Variant identity: gene, transcript, build, c.HGVS, p.HGVS, variant type

2. Clinical phenotype / suspected disease

3. Inheritance model and family structure

4. Parental genotype status and de novo / segregation details

5. Population / database / literature evidence

6. Functional and computational evidence

7. Criteria assignment and final review

At each step, summarize back in one compact block:

Safety / scope

Always say clearly:

Inputs you should collect

Use templates/intake.md and ask for or normalize these fields:

If transcript, genome build, or HGVS is unclear, stop and ask for clarification before classification.

Standard workflow

Step 1: Confirm scope

Proceed only if all are true:

1. Variant is a germline small variant (SNV/indel)

2. Naming/build/transcript are defined

3. User understands output is review-only

4. Any gene-specific ACMG framework has been checked

Step 2: Normalize the record

Create a clean variant record using templates/intake.md.

Step 3: Gather evidence by ACMG bucket

Pathogenic side:

Benign side:

Step 4: Assign criteria carefully

Use templates/evidence-table.md.

For each criterion, record:

Do not double count overlapping evidence.

Step 5: Evaluate conflicts

If both pathogenic and benign evidence exist:

1. Check whether evidence is truly independent

2. Downgrade/remove misapplied criteria if needed

3. If conflict remains unresolved, prefer VUS over forced certainty

4. State what additional data could resolve the conflict

Step 6: Apply combination logic

Use scripts/classifier.py or reproduce its logic manually.

Pathogenic if any:

Likely Pathogenic if any:

Benign if any:

Likely Benign if any:

Else: VUS

Guided questioning pattern

Use short, sequential prompts:

Included files