The Agent Skills Directory

[DATA_EXPOSURE]: The skill provides instructions for authenticating with OCI services by reading the user's local configuration file at ~/.oci/config. This is documented as a standard practice for OCI IAM authentication and is not associated with any exfiltration patterns.
[REMOTE_CODE_EXECUTION]: The skill documents the use of oci_adk, oci, fastapi, and slack_sdk. These are legitimate libraries used within their intended context for building and deploying AI agents. No unverified remote script execution patterns were detected.
[INDIRECT_PROMPT_INJECTION]: The skill describes patterns for processing external data from users (e.g., in a FastAPI endpoint or Slackbot). While this creates a potential surface for indirect prompt injection, it is the inherent function of a chat agent skill.
Ingestion points: chat_endpoint in SKILL.md (takes a message string) and handle_message in SKILL.md (takes text from a Slack event).
Boundary markers: None explicitly mentioned in the example code snippets.
Capability inventory: The agents utilize Agent.execute, Agent.send, and Workflow.execute to process input and potentially call tools like db_tool or crm_api.
Sanitization: No explicit sanitization or input validation logic is shown in the provided code examples.

Oracle ADK Expert