The Agent Skills Directory

[PROMPT_INJECTION]: The project includes common prompt injection strings (e.g., 'ignore all previous instructions') in files like references/testing.md and agents/test-writer.md. These are explicitly used as test cases for a security evaluation suite designed to ensure that agents are resistant to adversarial inputs, rather than attempting to subvert the behavior of the agent executing the skill.
[COMMAND_EXECUTION]: The skill is configured to use Bash and Write tools to perform its primary functions of project auditing and infrastructure scaffolding. It mitigates potential risks by documenting high-risk capabilities and mandating explicit user confirmation for any destructive or production-affecting operations.
[EXTERNAL_DOWNLOADS]: The skill references and scaffolds code using well-known, trusted libraries from established organizations such as the Vercel AI SDK, Anthropic SDK, and the Model Context Protocol (MCP) SDK. All referenced domains and repositories are official and verified.
[SAFE]: The overall architecture of the skill promotes defensive programming practices, including the use of Zod for schema validation, RFC 9457 for structured error recovery, and OAuth 2.1 for secure machine-to-machine authentication.

surface