Tencent scBert Single-Cell Pretraining Model

Problem

Single-cell pipelines often rely on a specific NEXTFLOW app on a hosted platform, but agent sessions can easily drift into out-of-scope commands, missing auth or project config, and name conflicts that break runs. This skill narrows the workflow to Tencent Health Omics Platform's scBert single-cell pretraining model for fine-grained annotation, novel subpopulation discovery, and Marker selection.

How It Works

• Command scoping: It builds commands through scripts/omics_cli.py, with a locked AppId, NEXTFLOW app type, and public-app parameters.
• Prechecks: It checks whoami and config show; trial users also receive quota checks based on remaining days and today's run budget.
• Name protection: Before import, it enforces a same-name check using --type NEXTFLOW, and asks for a new name on conflict instead of overwriting.
• Run constraints: It requires selecting nf_version, uses the default InputTemplate, and allows --input overrides; local file inputs must be uploaded to COS first.

Boundaries

• App-only scope: It cannot run other applications, project-owned apps, or substitute another AppId.
• Failure handling: Out-of-scope commands, auth failures, or missing platform configuration should be handled by the skill's guidance, omics login, or the platform setup flow.