Reviewing AI-Generated ABAP, CAP and Integration Code: Quality Gates with ATC, ABAP Unit and Mapping Tests

Keep this: job aid

Quality gates for AI-written SAP code

• Review intent, authorization, test meaning, suppressions and scope; let gates handle validity.
• ABAP: syntax clean, ATC with no priority 1 or 2 findings at transport release, ABAP Unit green, 80% statement coverage on changed classes.
• Agents may run abap_object_check, abap_run_atc and abap_run_unit_tests; deny exemption requests and transport release.
• Break the behavior once: at least one generated test must fail.
• CAP: cds lint clean; cds.test with allowed and denied users for every authorization rule.
• UI5: ui5lint --format json in the pipeline; no deprecated APIs or globals.
• Integration: simulate mappings, simulate flows with mocked receivers, and run payload regression on golden pairs with normalization.
• Hooks: PostToolUse runs lint, Stop runs tests; protect existing test files; CI reruns every gate.
• Tag AI-assisted changes on transports and pull requests; track first-pass rate, suppressions and escaped defects.
• Agents work in CMD and cmc-ext-dev only; releases stay with people.