Tencent CD-GPT Generative Bio Foundation Model

Problem it solves

When engineers run CD-GPT Collection (Nextflow) on the Tencent Health Omics Platform, the friction is usually operational rather than scientific: the collection AppId cannot be submitted directly with omics run, sub-app AppIds must be exact, trial quotas may block execution, duplicate project app names can trigger DUPLICATE_APP_NAME, and locally generated files must be uploaded to COS before the run container can read them. This skill narrows those constraints into a fixed workflow and prevents out-of-scope CLI calls.

How the skill works

It only targets two Nextflow sub-apps in CD-GPT Collection (Nextflow): translation generation and reverse translation generation. The workflow covers:
- CLI and auth checks: verifies omics is available, runs whoami and config show, and guides users through login or Region / ProjectId / EnvironmentId / CosBucketName setup when needed.
- Sub-app matching: selects from the hard-coded list, such as CD-GPT_translation_generation or CD-GPT_reverse_translation_generation, without using the collection AppId.
- Pre-run confirmation: checks duplicate app names, confirms --public-app-name, --nf-version, and --input, and enforces quota checks for C-end users.
- Execution and diagnostics: invokes omics run, handles orphan applications, missing NF versions, parameter merge failures, or auth failures, and reminds users to inspect the outdir when present.

Boundary

The skill is scoped to CD-GPT translation and generation workflows for DNA, RNA, and protein sequence tasks. It is not a general omics task skill and cannot run applications outside this collection. When local files are inputs, users must specify and confirm the COS upload path.