ultr-erp-dump-import

name	ultr-erp-dump-import
description	Use when asked to "/UltrERP-dump-import", "import a dump", "dump import", "extract a SQL dump", or stage archival legacy CSV exports. This skill is for archival or break-glass dump workflows, not the routine live refresh path.
argument-hint	[phase] --batch-id <batch-id> [--schema raw_legacy] [--tenant-id <uuid>]

Use this skill when the operator explicitly wants to run or plan the reviewed dump-based legacy-import workflow, starting from a raw SQL dump file or from already-extracted CSV files. Do not use it for routine live legacy DB refreshes; those should go through the reviewed live refresh and promotion surfaces.

This skill is an orchestration wrapper around the backend CLI. Do not reimplement import business logic in Markdown, Python snippets, or ad hoc shell pipelines. Always route workflow execution through the reviewed CLI in the command map.

Use these operating rules:

1. Start by identifying the requested phase, batch id, schema, tenant id, and any file path arguments.
2. Use the stable backend invocation path from the command map: run commands from backend/ with uv run python -m domains.legacy_import.cli ....
3. On fresh environments or when the CLI reports missing legacy-import control tables, stop and ask the operator to bring the schema to the current Alembic target before retrying. Use the repo-documented upgrade head path first, and if Alembic reports multiple heads on the active branch, ask for confirmation before using the upgrade heads fallback documented in the safety and validation guide.
4. Keep shell approval conservative. This skill intentionally does not pre-approve shell or bash. Follow the confirmation policy in the safety and validation guide.
5. Treat extract, stage, normalize, map-products, export-product-review, import-product-review, canonical-import, and validate-import as commands that write files or persistent state. Before running them, show the exact command, input path, output directory, batch scope, schema, tenant, and file paths, then wait for explicit operator confirmation.
6. Prefer explicit scope arguments over hidden defaults. If the operator does not provide a schema, tenant id, source directory, review CSV path, or approval identity that matters for the requested phase, ask for it instead of guessing.
7. After every CLI run, report the exact subcommand, exit status, batch id (where applicable), schema, tenant id (where applicable), and any attempt number that was used or returned.
8. For validate-import, do not rely on terminal prose alone. Capture the json= artifact path from CLI output, open that file, and summarize the machine-readable fields described in the safety and validation guide. A blocked validation can legitimately exit with status 1, so treat the artifact as authoritative when it is present.
9. If the operator wants this workflow moved under erp-skills/, explain that erp-skills/ is not a default project-skill discovery path. The supported locations are .github/skills/, .Codex/skills/, and .agents/skills/ unless chat.skillsLocations is configured explicitly.
10. If validate-import reports that no canonical import run exists for the batch, stop and explain that validation only works after a completed canonical-import attempt for that same batch.
11. extract is the only phase that reads a raw SQL dump and writes files. It is idempotent — re-running overwrites the output directory. It is also the only phase that has no database dependency.

Smart defaults for extract

When the operator provides a SQL dump path, auto-derive parameters without asking:

Parameter	Derivation
--batch-id	Stem of the input filename (e.g. cao50001.sql → cao50001)
--output	Ask for an explicit archival path; do not default to legacy-migration-pipeline/extracted_data
--format	csv (the only format consumed by the import pipeline)
--schema	raw_legacy (always correct)
--tenant-id	00000000-0000-0000-0000-000000000001 (DEFAULT_TENANT_ID)

Only auto-derive --batch-id, --format, --schema, and --tenant-id. Always ask for an explicit archival --output path.

Suggested execution flow

1. Read the requested phase in the command map. extract (phase 0) is the entry point when starting from a raw SQL dump; stage (phase 1) is the entry point when archival CSV files already exist.
2. If the operator needs syntax or available flags, run the corresponding --help command first.
3. If the phase writes data or files, apply the confirmation checklist from the safety and validation guide.
4. Run the reviewed CLI command.
5. If validation was requested, open the JSON artifact and summarize blockers, replay metadata, and stage discrepancies from the artifact itself.

Use this skill for legacy-import workflow guidance only. Do not use it for unrelated ERP tasks, generic data migration advice, or non-CLI import implementations.