This is the Inception phase. Output is an assessment and a migration plan the human approves before any conversion happens. Mirrors AI-DLC Inception: gather requirements, analyze, propose a plan, stop for approval.
Active engine pair
Determine the pair from the connection engines: <source>-to-<target> (e.g.
oracle-to-postgresql, sqlserver-to-mysql). Substitute it for <pair> below — the engine
definition is engines/<pair>/ and the knowledge base is skills/<pair>-playbook/.
Inputs
connections.yaml,migration-config.yaml(resolved by the orchestrator)- Active engine pair definition
engines/<pair>/engine.yaml - Playbook reference skill
<pair>-playbookfor compatibility lookups
Steps
1. Re-verify pre-flight (idempotent)
Ensure package deps are installed (pip install -r scripts/requirements.txt), then run
python -m dbmig test-connection --side both. Record driver versions and connectivity
status into 01-assessment/preflight.md. Stop if it fails. (No native client tools are
used — connectivity is via Python drivers.)
2. Inventory the source
For each schema in migration-config.yaml scope.schemas, run:
python -m dbmig inventory --schema <SCHEMA> --project <project>
This writes structured 01-assessment/inventory.yaml and inventory.json (object counts,
tables + row counts, datatypes in use, stored-code volume). Summarize it into a clean
01-assessment/inventory.md table.
3. Compatibility assessment
For each object type and notable construct found, consult the active pair's playbook
references (skills/<pair>-playbook/references/...) and classify each item:
| Category | Meaning |
|---|---|
| Automatic | Direct equivalent; low risk (most tables, standard types, basic DML) |
| Assisted | Mechanical rewrite with a known pattern |
| Manual | Needs redesign / human judgment |
| Blocked | No supported path yet; flag for the user (source-only features) |
Use engines/<pair>/datatype-map.yaml for datatype risk and the pair's SQL/code references
for construct-level guidance (each topic file carries a conversion category). Cite the
specific reference file per finding so it traces to guidance.
4. Effort & risk
Summarize: object counts by category, stored-code line counts, top risks, and the known
semantic differences that will need explicit testing for this pair — see
engines/<pair>/checks/equivalence-spec.md (e.g. empty-string/NULL, date arithmetic,
numeric rounding, case sensitivity, IDENTITY/sequence handling).
5. Migration plan
Write 01-assessment/migration-plan.md with:
- Scope (schemas/objects in and out).
- Conversion order (tables → constraints/indexes → sequences → views → functions → procedures/packages → triggers), grouped into units of work that can be converted and tested independently (AI-DLC parallel construction principle).
- Testing approach: data path (
frameworkvsdms), and which equivalence checks apply. - Cutover approach (high level; detailed in Operations).
- Open questions / decisions needed from the user.
Gate
Present migration-plan.md and the compatibility summary. Stop and ask the orchestrator's
gate question: "Approve the assessment & migration plan?" Do not start conversion. If the
user wants changes (scope, ordering, strategy), revise the plan and re-present.
Notes
- Read-only on the source. This phase must not modify the source or target.
- Be honest about Blocked/Manual items — surfacing them now is the point of Inception.