The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves object names, packages, and other metadata from the SAP system. This creates a surface for indirect prompt injection if an attacker names SAP objects with malicious instructions intended to override agent behavior when the metadata is processed.
Ingestion points: The agent incorporates results from tools such as SAPQuery, SAPSearch, and SAPRead (which fetch SAP object and usage data) into its context.
Boundary markers: There are no instructions to use delimiters or ignore-instruction warnings around the retrieved data.
Capability inventory: The skill utilizes several SAP-specific tools, including SAPQuery, SAPRead, SAPSearch, and SAPNavigate.
Sanitization: The skill does not specify any sanitization or escaping of retrieved SAP metadata before it is presented in reports or used in subsequent steps.
[DYNAMIC_EXECUTION]: The skill provides instructions for the agent to dynamically construct SQL strings for the SAPQuery tool using user-supplied inputs like package names or namespace patterns.
Evidence: Step 3 and Step 1 describe building WHERE clauses by interpolating object names or search patterns directly into SQL templates.
Risk: This pattern introduces a potential for SQL injection if the agent does not properly escape or validate user inputs before placing them into the SQL string.
[EXTERNAL_DOWNLOADS]: The skill provides a link to an external resource for parsing specific data types that the skill itself does not support.
Source: Points to a script in a repository under the official AWS solutions library samples organization on GitHub.
Note: This reference targets a well-known service and is consistent with the skill's intended development and auditing purpose.

sap-unused-code