| name | sap-code-agents-analysis |
| description | Evidence-backed analysis of custom ABAP codebases, including object search, dependency and where-used analysis, usage evidence, migration findings, retirement candidates, and technical documentation. Adapted from jonathanbernsteiner/sap-code-agents. |
| trigger | {"keywords":["sap code agents","analyze abap codebase","abap evidence","migration assessment","retirement candidates","usage statistics","dependency graph"],"intent":"Analyze an ingested or accessible custom ABAP codebase with traceable evidence for every claim"} |
SAP Code Agents Analysis
Use the evidence-first method adapted from jonathanbernsteiner/sap-code-agents.
Routing
- Discover objects with the registered ADT/code-search capability.
- Read the complete source of every object used as evidence.
- Resolve callers and dependencies with where-used and repository search.
- Use runtime usage data only when an authorized source is available; otherwise label it unavailable.
- Produce findings that cite object names and source line evidence. Never convert an inference into a fact.
For a local abapGit checkout, search and read the files directly. For a live system, prefer the router's ADT providers. Do not ask for credentials in chat and do not bypass the configured MCP approval policy.
Deliverables
- Functional purpose per object, supported by retrieved source.
- Dependency and data-access map.
- Migration risks with explicit evidence tiers.
- Retirement candidates separated into
supported, inferred, and unresolved.
- Open questions for missing source or usage telemetry.
Evidence rules
- Attach at least one verifiable object/source reference to each material claim.
- Report parser or retrieval failures instead of silently omitting objects.
- Keep system/workspace scope on every query; never mix evidence across systems.
- Do not claim that unused code is safe to retire from static analysis alone.
Upstream reference
The pinned snapshot is cataloged as jonathanbernsteiner-sap-code-agents under bundled/skills/. Its architecture and evidence model are described in its README.md; its application is not an MCP server.
Verification
Confirm that object counts reconcile, all cited objects were retrieved, unresolved items are listed, and conclusions can be traced to source or authorized usage evidence.
