The Agent Skills Directory

[SAFE]: The skill serves as an educational and architectural reference for multi-agent design patterns such as Orchestrator-Workers, ReAct, and Supervisor patterns.
[PROMPT_INJECTION]: No override instructions, safety bypasses, or malicious role-play prompts were identified. The skill includes a section on 'Conversation Control' which explicitly provides patterns for implementing guardrails.
[DATA_EXFILTRATION]: No network operations (curl, wget) or access to sensitive file paths (e.g., .ssh, .aws) were found. References to the 'memory' MCP server are standard for the intended agent environment.
[REMOTE_CODE_EXECUTION]: The skill contains illustrative Python code snippets for architecture patterns. These examples do not involve remote script execution, unpinned dependencies, or unsafe package installations.
[DYNAMIC_EXECUTION]: The code examples provided are static and do not use dangerous functions like eval() or exec() on untrusted input. They use standard Python libraries for state management and coordination.
[INDIRECT_PROMPT_INJECTION]: While the skill defines patterns for processing user-supplied tasks and goals, it also documents mitigation strategies including 'Behavioral Guardrails' and 'Conflict Resolution' to handle potentially adversarial data in multi-agent workflows.

multi-agent-patterns