The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill represents a vulnerability surface as it processes external OData metadata files (metadata.xml) and user-provided definitions to generate or modify application code.
Ingestion points: The agent is instructed to fetch and read OData service metadata from external backends or local files (/webapp/localService/mainService/metadata.xml) in SKILL.md.
Boundary markers: There are no explicit instructions to sanitize or ignore embedded instructions within the processed metadata.
Capability inventory: The skill allows modification of manifest.json, creation of UI components, and execution of shell commands (npm start, npm run watch).
Sanitization: No sanitization or validation of the ingested metadata structure is specified.
[COMMAND_EXECUTION]: The guidelines instruct the agent to execute various shell commands to preview and manage the application.
Evidence: Mentions execution of npm start, npm run start-mock, and project-specific watch scripts (e.g., npm run watch-manage-travel).
[DYNAMIC_EXECUTION]: The skill involves the dynamic modification and assembly of application configuration and UI logic at runtime.
Evidence: Instructions for modifying manifest.json data sources and creating frontend annotations in annotation.xml based on entity definitions.
[PROMPT_INJECTION]: The skill contains an instruction that overrides standard agent verification behavior, potentially reducing oversight.
Evidence: "After the Fiori MCP Server tools execute successfully, inform the user of completion. Do not verify or double-check
treat successful tool execution as confirmation."

sap-fiori-app-development