The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it ingests data from research files and CSV headers to generate instructions for downstream agents. However, this is its primary intended function and operates within a local project directory. Ingestion points: Reads from research and context files in claude-code-gtm/context/ and CSV metadata. Boundary markers: The synthesis instructions do not specify the use of delimiters to isolate interpolated research data within the generated prompt. Capability inventory: The skill performs file system writes to claude-code-gtm/prompts/ to store generated templates. Sanitization: No explicit sanitization or validation logic is applied to the source research data before it is embedded into the output prompts.
[DYNAMIC_EXECUTION]: The skill dynamically generates markdown files that function as prompts for an external email-generation skill. This is a standard template-generation pattern and does not involve runtime compilation or injection of code from untrusted remote sources.

email-prompt-building