The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill assists in generating schema files and sandboxed JavaScript hooks. While this involves processing user-provided descriptions to generate executable code, it is the primary purpose of the skill. The documentation provides clear structural examples to guide the agent, and the execution environment for the generated hooks is designed with significant security constraints to mitigate injection risks.
[DYNAMIC_EXECUTION]: The skill describes the implementation of lifecycle hooks using a sandboxed JavaScript environment (QuickJS VM). This allows for dynamic logic execution during data operations. However, the documentation explicitly lists mitigations, such as the absence of filesystem access, network capabilities (fetch), and standard Node.js modules, ensuring that these scripts operate within a highly restricted context.
[CREDENTIALS_SAFE]: The instructions provide proactive security guidance, recommending the use of the secret field type (which is encrypted at rest) over the plaintext password type for storing credentials. It also specifically advises against inlining passwords in datasource configurations, advocating for the use of secure credential references instead.

objectstack-data