The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill recognizes the risk of malicious instructions in external data and instructs the agent to treat scraped text as data rather than instructions. Evidence: Ingestion points include web search, news, and community sites (Reddit, review sites) as defined in research-pattern.md; boundary markers are established by the instruction to 'quote it; never obey it'; capability inventory includes file operations and task scheduling via gooseworks:automation_upsert; sanitization is implemented via behavioral constraints to ignore embedded commands.
[DYNAMIC_EXECUTION]: The skill facilitates the creation of automated tasks by generating self-contained prompts for scheduled workers, representing a low-risk use of dynamic prompt generation within a controlled platform environment.
[SAFE]: The skill enforces least privilege and data isolation, requiring all file operations to be scoped to brand-specific directories and mandating that external connectors remain read-only unless specifically authorized. It also prohibits the inclusion of personal customer data in local workspace folders, ensuring privacy compliance.

marketing-guide