The Agent Skills Directory

[SAFE]: The skill is entirely documentation-based, providing an architectural guide and file tree structure for building AI agents. It does not include any executable scripts or active code that could pose a runtime risk.\n- [NO_CODE]: All content consists of Markdown-based guidelines and illustrative TypeScript snippets. There are no files provided that the agent is instructed to execute directly.\n- [COMMAND_EXECUTION]: The documentation references developer-oriented command-line tools, such as npx @ai-sdk/codemod v7, for code migration. These are intended for human developers setting up the environment and are not automated by the skill itself.\n- [INDIRECT_PROMPT_INJECTION]: The guide explicitly addresses how to handle untrusted input by recommending the use of tagged sections for static instructions and enforcing strict schema validation for all tool inputs, which serves as a mitigation strategy against indirect injection attacks.\n- [DATA_EXPOSURE]: The architecture advocates for security boundaries by isolating sensitive operations (marked with .server.ts) and ensuring that tool contexts only expose data explicitly required by their validated schemas.

agent-codebase