The Agent Skills Directory

[DATA_EXFILTRATION]: The skill's code examples demonstrate accessing sensitive credential files located at ~/.oci/config. Accessing system-level configuration and credential stores is a sensitive operation that exposes environment metadata and authentication secrets to the agent context.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates an environment where indirect prompt injection is possible. It instructs the agent on how to build applications that ingest untrusted user input (e.g., via the chat_endpoint or user_query variables) and pass that data to LLM agents which have access to enterprise tools such as SQLTool and FunctionTool.
Ingestion points: Untrusted data enters the agent context through the user_query parameter in orchestration logic, user_input in workflows, and the message field in the FastAPI and Slack integrations (SKILL.md).
Boundary markers: The provided implementation examples lack explicit delimiters (e.g., XML tags or clear text separators) or instructions to ignore embedded commands within user-provided data.
Capability inventory: The agents described in the skill are granted capabilities to perform database operations via SQLTool, interact with CRM systems through FunctionTool, and invoke other specialized agents.
Sanitization: There is no evidence of input validation, sanitization, or filtering logic in the code snippets to prevent malicious instructions hidden in user data from being executed by the LLM.
[EXTERNAL_DOWNLOADS]: The skill references official documentation and resources from oraclecloud.com and oracle.com. These are recognized as well-known service domains and the references are documented here neutrally as safe project resources.

Oracle ADK Expert