The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides prompt templates designed to ingest untrusted external data, such as customer email replies and web research findings. This creates a surface for indirect prompt injection. However, the skill explicitly provides mitigation strategies, including strict output schemas (JSON), requirements for source citations, and instructions to avoid guessing when information is missing. Ingestion points: Variables in references/prompt-library.md such as {{reply_body}} and {{signal}} where untrusted user or web content is interpolated. Boundary markers: Prompt templates enforce structured JSON outputs to separate data from agent instructions. Capability inventory: The templates are intended for use with external GTM tools like Claygent for web research and n8n for workflow automation. Sanitization: The skill mandates design rules including word limits, banned phrases, and explicit 'do not guess' constraints to maintain output integrity.
[COMMAND_EXECUTION]: The script scripts/check-output.py is a local utility used to validate that prompt specification files contain required markdown headers. It reads file content based on user-provided paths but does not execute the content or perform any network operations.

ai-prompts-toolkit