The Agent Skills Directory

[PROMPT_INJECTION]: A static detector flagged an 'ignore instructions' pattern in references/official-docs/integrations/ag-ui/security-considerations.md. Analysis confirms this is a false positive; the string is part of a security tutorial explaining how to prevent such attacks.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it is an orchestration framework for LLM agents.
Ingestion points: User prompts and external data inputs are processed across various executors described in SKILL.md and AGENT.md.
Boundary markers: Present; documentation repeatedly advises treating MCP and A2A services as trust boundaries and using explicit approval steps for tools.
Capability inventory: The routing logic in AGENT.md supports tools including Bash, and the framework supports network and file system access via MCP as described in mcp.md.
Sanitization: Explicit guidance on input validation and sanitization is provided throughout the reference documentation, particularly in the security-considerations.md file.

dotnet-microsoft-agent-framework