migrate-custom-code

Warn

Audited by Gen Agent Trust Hub on Jun 24, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill defines 'Smart Defaults' with instructions for the agent to 'apply silently, do NOT ask' the user when auto-detecting object types or selecting check variants. This overrides standard agent behavior and reduces user control over the analysis phase.
  • [DATA_EXPOSURE_AND_EXFILTRATION]: The skill reads full ABAP source code from the SAP system and transmits metadata, including ATC finding titles and deprecated API names, to external services via the search and sap_notes_search tools to retrieve migration guidance.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external searches and system findings to generate code fixes, creating an injection surface.
  • Ingestion points: Findings from SAPDiagnose, documentation from search(), and SAP Notes from sap_notes_search().
  • Boundary markers: The instructions do not specify markers to isolate retrieved data from agent logic.
  • Capability inventory: The skill has broad system access, including SAPWrite (edit_method, update) and SAPActivate functions.
  • Sanitization: No validation or sanitization of search results is mentioned before they are used to generate ABAP code proposals.
  • [DYNAMIC_EXECUTION]: The skill generates executable ABAP code at runtime and writes it to the target system. It also utilizes the apply_quickfix action, which processes and applies dynamic code deltas to system objects.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jun 24, 2026, 08:00 AM
Security Audit — agent-trust-hub — migrate-custom-code