The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided natural language descriptions to generate security risk policies, which introduces a potential surface for indirect prompt injection. This risk is effectively mitigated by multiple mandatory manual review checkpoints: the agent is instructed to confirm the organization, establish the exact project slug, review policy drafts with the administrator, and obtain explicit confirmation before performing any write operations.
[DYNAMIC_EXECUTION]: The skill utilizes platform-specific tools (execute, call_platform_tool) to dynamically discover and invoke other platform capabilities at runtime. This dynamic tool resolution is restricted to the platform's MCP environment and is governed by access checks, with the skill specifically requiring the agent to stop for user confirmation before executing any tool that performs a write operation.

create-guardrail-from-description