The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes Azure resource metadata and change logs which could contain malicious strings intended to influence the agent's behavior.
Ingestion points: Data is retrieved from Azure Resource Graph via execute_query against the resourcechanges and Resources tables (referenced in SKILL.md).
Boundary markers: There are no explicit instructions or delimiters used to separate the ingested resource data from the agent's instructions.
Capability inventory: The skill uses execute_query, query_costs, and generate_query to interact with Azure resources and billing data.
Sanitization: No sanitization or filtering of resource tags or metadata is mentioned before the data is summarized by the agent.
[SAFE]: The skill is authored by a trusted vendor and its capabilities are well-aligned with its intended purpose of cost analysis. It incorporates security best practices by explicitly instructing the agent to pass all generated queries through a validation tool (validate_query) before execution. No patterns of data exfiltration, unauthorized persistence, or malicious code execution were detected.

cost-investigation