The Agent Skills Directory

[SAFE]: The skill provides documentation and implementation patterns for a legitimate cloud integration between Microsoft Foundry hosted agents and CopilotKit frontends. It adheres to standard development workflows and security best practices.
[EXTERNAL_DOWNLOADS]: The skill references standard packages from official registries such as npm and PyPI. These include @copilotkit/runtime, agent-framework-foundry, and azure-identity, which are well-known dependencies for the described architecture and originate from established sources.
[COMMAND_EXECUTION]: The documentation includes standard development commands using the azd (Azure Developer CLI) and curl for testing API endpoints. These commands are used appropriately for local testing, deployment, and debugging purposes.
[INDIRECT_PROMPT_INJECTION]: The skill proactively addresses potential injection and autonomous execution risks by recommending Human-in-the-loop (HITL) approvals for side-effecting tools (e.g., using approval_mode="always_require"). It also suggests architectural safeguards such as avoiding the inlining of large documents directly into prompts.
[CREDENTIALS_UNSAFE]: References to authentication mechanisms follow industry-standard security patterns. The skill demonstrates the use of DefaultAzureCredential and Entra ID tokens, correctly pointing toward environment variables and managed identities rather than hardcoding sensitive secrets.

foundry-hosted-agent-copilotkit