Clean Core

Classify a package against the SAP Clean Core compliance levels, prepare per-object fix context, drive the fix with an AI agent, and apply the result.

SAP Clean Core decouples custom extensions from SAP standard code so S/4HANA upgrades stay safe. Custom code that depends on internal or non-released SAP APIs creates upgrade risk; the Clean Core Extensibility whitepaper defines four compliance levels that quantify it. abapctl runs the assessment, prepares the fix context, and applies the fixes; the fix step itself is where an AI agent does the work and abapctl acts as the live SAP-side oracle.

For driving the loop with an agent, see Skills & Agents and Working with AI agents. To walk it end to end, see the Clean Core use case.

1. Assess

abapctl clean-core assess ZFINANCE -c s4h

Discovers every object in the package, runs ATC under the configured Clean Core variant (clean_core.atc_variant, default CLEAN_CORE), and classifies each object:

Level	ATC result	What it means	Action
A	No findings	Fully compliant with Clean Core	Cloud-ready, no action needed
B	Info only	Uses documented extension points	Acceptable, low upgrade risk
C	Warnings	Uses internal or undocumented APIs	Conditionally clean. Verify before upgrades.
D	Errors	Non-released APIs, modifications, or blocked patterns	Requires remediation

Output goes to clean-core/{SID}/{PACKAGE}/:

• reports/SUMMARY.md: per-object table with level and finding count
• reports/<OBJECT>_atc.md: per-object Markdown report
• state/discovery.json: full object inventory
• state/progress.json: resumable checkpoint
• state/<OBJECT>_findings.json: raw structured findings

The run is idempotent and resumable. Re-running picks up where it left off; pass --force to start fresh. Exit code 0 means clean (A/B), 1 means findings exist (C/D), 2 means error. For a cross-package roll-up:

abapctl clean-core executive

2. Prep

abapctl clean-core prep S4H/ZFINANCE

For each D and C object, downloads the source plus per-finding fix context. Output lands under clean-core/{SID}/{PACKAGE}/fix-context/:

• <obj>.<type>.abap: editable source (abapGit/AFF naming)
• .context/<obj>.findings.json: structured findings (line, message, deprecated API, suggested successor, doc URL)
• .context/<obj>.context.md: human-readable fix brief

3. Fix (AI agent loop)

Example flow. Adapt to your agent's capabilities.

For each priority-1 finding on a Level D object, the agent searches for a released-API replacement using three sources, stopping at the first hit:

1. ATC tags on the finding. Every ATC finding ships per-finding metadata (refObjectName, refObjectType, Successors: info). When the SAP system has already classified the deprecated API and its successor, that data is right there in the finding.
2. SAP Cloudification Library reference JSONs. Run abapctl reference update once, and objectReleaseInfoLatest.json + objectClassifications_SAP.json land under .abapctl/reference/sap-api-reference/. These cover the broad catalog of deprecated APIs and their cloud-ready successors.
3. Live CDS view query. When the first two are inconclusive, abapctl release-state lookup <api> queries I_APISWITHCLOUDDEVSUCCESSOR on the connected SAP system directly. Always current to whatever release the target system is on.

When the agent reports the object as clean, the human reviews the diff, then runs apply.

4. Apply

abapctl clean-core apply S4H/ZFINANCE --dry-run    # preview every write
abapctl clean-core apply S4H/ZFINANCE              # commit

Locks each object, uploads the fixed source, unlocks, and activates. On failure the lock-finally guard releases the lock rather than leaving the object stuck, and you re-run to continue from the failure point.

See also

• Skills & Agents and Working with AI agents: drive the fix loop.
• Clean Core use case: a worked example that runs this loop end to end.
• Quality checks: the ATC mechanics underneath assessment.