The Agent Skills Directory

[SAFE]: The skill demonstrates a strong commitment to security, incorporating advanced patterns for tool sequencing, role-based access control, and policy-driven execution. No malicious patterns such as prompt injection, obfuscation, or unauthorized exfiltration were detected.\n- [INDIRECT_PROMPT_INJECTION]: While the skill interacts with external tool outputs, it treats them as untrusted data. It mitigates injection risks by enforcing Model Context Protocol (MCP) tool contracts, using inter-agent communication schemas, and mandating input/output validation.\n- [COMMAND_EXECUTION]: The skill includes the execute_command tool but wraps its use in strict safety protocols. It requires identifying active roles, checking actions against action-boundaries.yaml, and obtaining explicit user approval for state-changing or destructive commands.\n- [DATA_EXFILTRATION]: Comprehensive data classification and redaction strategies are integrated via data-classification.yaml to ensure that sensitive identifiers, PII, and credentials are not logged or transmitted in handoff artifacts.

agent-tool-orchestration