The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to process untrusted external data, such as tools/list responses, source files, or OpenAPI definitions. While this creates a potential ingestion point for malicious instructions, the risk is minimal because the skill is limited to analysis and reporting.
Ingestion Points: Untrusted data enters the agent's context through pasted tool definitions or OpenAPI files as specified in the review instructions.
Boundary Markers: The instructions do not define specific delimiters for the external content, though the agent is instructed to focus strictly on technical review and citation.
Capability Inventory: The skill does not request access to any system tools, network operations, or file-writing capabilities, ensuring that any embedded instructions in the processed data cannot trigger harmful actions.
Sanitization: No specific sanitization or escaping of the input data is mentioned, relying instead on the agent's instructions to treat the input as an artifact for review.
Trusted External References: All external links provided in the skill instructions point to official Microsoft documentation (learn.microsoft.com), which are recognized as trusted sources. These references are used to provide evidence-based verdicts for the compatibility checks.
Operational Safeguards: The skill includes explicit guardrails that prevent the agent from attempting to connect to the reviewed servers or modifying the user's tenant, ensuring the audit remains static and non-intrusive.

copilot-studio-mcp-readiness