On-demand BioLine downstream Tools
Downstream execution selects an exact BioLine Tool version supplied for the task. The five image entries below are retained as migration inventory from the earlier OpenClaw workspace; they are not a runtime allowlist and are never selected by the execution code.
| historical key | historical image tag | former network | former scope |
|---|---|---|---|
downstream | openclaw/downstream:1.1.1 | none | Python/R statistics, machine learning, microbiome analysis, reports |
downstream_dl | openclaw/downstream-dl:1.0.0 | none | downstream tasks requiring deep-learning runtime |
amplicon | openclaw/amplicon:1.0.0 | none | amplicon-related downstream processing |
base | openclaw/base:1.0.0 | none | lightweight text, configuration, and adaptation |
isolate_ont_kofam_cns | openclaw/isolate-ont-kofam-cns:1.1.0 | none | isolated ONT, Kofam, and CNS runtime |
Prepare an exact Tool version
After upstream finalization, call prepare_downstream with the exact NAME:VERSION. OpenClaw exports that version's summary and config before pulling it. The summary must declare analysis_lane: downstream and fixed_command: false; the Tool config must have no fixed command. A missing or mismatched export stops preparation before the pull. OpenClaw then calls fkit pull NAME:VERSION and derives the local tag as name.lower():version. Any exact, positive integer version discovered in BioLine can be used; no local catalog entry is required.
Container boundary and traceability
Only finalized upstream output enters downstream. Mount stage/ and task reproducibility/ read-only, keep the network disabled, and make only analysis/ writable. The container carries no authoring labels and never reuses an authoring build container. Persist the exact Tool reference, name, version, summary hash, derived tag, and pre/post image identities in the Job registry. Keep the generated task script, command, outputs, logs, and exit status in the Job workspace.