aws-aidp-migrator

# AWS → Oracle AIDP migrator This skill drives the `aws-aidp` CLI, which migrates an AWS data stack to Oracle AIDP in four verbs. The translators are **deterministic-first**: each rewrite is a named rule with a reason, and anything that can't be safely rewritten is **flagged and left in place**, never silently changed. Preserve that principle — never hand-edit a flagged construct into a silent rewrite. ## Prerequisite The CLI must be installed: ```bash pip install -e . # from the plugin directory pip install -e "${CLAUDE_PLUGIN_ROOT}" # or, when installed as a plugin ``` There is deliberately no PyPI fallback here: `aws-aidp-migrator` is not a published package, and an unregistered name in instructions an agent executes is a name-squatting risk rather than a convenience. Live AWS mode uses the standard boto3 auth chain (`AWS_PROFILE`, `~/.aws/credentials`, IAM role). Offline/demo mode needs no AWS account. ## The four verbs (run in order) ```bash aws-aidp inventory --region us-east-1 -o inv.json # read-only AWS scan → manifest aws-aidp plan inv.json -o plan.json # manifest → AIDP mapping plan aws-aidp migrate plan.json --demo -o migrated # translate; --demo = offline artifacts aws-aidp verify migrated # classify: PASS / REVIEW / SKIP / FAIL ``` Fixture mode (no AWS): `aws-aidp inventory --fixture demo -o inv.json`. ## How to help the user 1. **Confirm scope first** — which AWS sources (s3, glue, athena, emr, sagemaker) and which region. Use `--sources` / `--filter` to narrow when they only want a slice. 2. **Always inventory → plan → migrate → verify in that order.** Each verb consumes the previous one's output file. 3. **Default to `--demo`** for `migrate` unless the user explicitly asks for a live AIDP write — the live write-side is not GA yet (roadmap v0.5). 4. **Read the reports.** After `migrate`, open `migrated/report.md` and surface the diffs and any `flag` findings — those need a human. After `verify`, report the PASS/REVIEW/SKIP/FAIL counts and explain SKIP = translator not built yet (EMR/SageMaker are stubs). 5. **Translating a single script/query?** You can call the translators directly: `python3 -c "from aws_aidp.translate.athena_to_spark_sql import translate; ..."` or `aws_aidp.translate.glue_to_spark`. ## Coverage today (be honest about it) - ✅ Athena SQL → Spark SQL, and Glue ETL → PySpark — translated to reviewable artifacts on disk. - ✅ S3 buckets → a generated `rclone` transfer job. Review it before running; the tool never moves data itself. - 🚧 Glue **Data Catalog** (databases, tables) — inventoried and planned, but no DDL is emitted yet, so `verify` reports every catalog asset as SKIP. - 🚧 EMR, SageMaker translators — stubs (roadmap). `verify` reports them as SKIP. - 🚧 Live AIDP write-side — nothing is written to AIDP. Applying the artifacts is manual today. See `references/verbs.md` for flags, and the plugin README (`${CLAUDE_PLUGIN_ROOT}/README.md`) for the full rule tables.