| name | cvm-operator |
| description | Operate the CVM pipeline end-to-end with reproducible commands. Use when asked to onboard CSV drops, run churn/cross_sell/reactivation scoring, produce top-N actionable customers, or deliver run artifacts and KPI summaries. |
CVM Operator
Run the pipeline through deterministic scripts, not ad-hoc command sequences.
Workflow
- Select input mode.
- Use
real mode when the user provides a client CSV folder and mapping JSON.
- Use
demo mode when no client data is available.
- Run the orchestrator script.
- From repo root:
bash skills/cvm-operator/scripts/run_cvm_flow.sh --mode <real|demo> --use-case <churn|cross_sell|reactivation> --top-n 200
- For all use-cases in one run:
bash skills/cvm-operator/scripts/run_cvm_flow.sh --mode <real|demo> --all-use-cases --top-n 200
- Return fixed deliverables.
- Pipeline summary JSON from
outputs/latest_run_summary*.json
- Top customers CSV from
outputs/<use_case>_top<N>.csv
- If
real mode: mapping report from outputs/client_mapped_raw/mapping_report.json
Required Inputs
For real mode, require:
--client-dir path--mapping-file path
For demo mode, optional:
--raw-dir path (if omitted, synthetic data is generated automatically)
Output Contract
Follow references/output-contract.md.
Notes
- Keep
CVM_ENABLE_FEAST=0 and CVM_ENABLE_MLFLOW=0 by default for fast local runs.
- Keep
CVM_ENABLE_SIGNALS=1 by default so explanations are meaningful.
- If reason quality gate fails, rerun once with
CVM_SIGNALS_REQUIRE_REASON_QUALITY_GATE=0 and report that override explicitly.
