The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill workflow is designed to load brand voice and persona context from local files, which serves as an ingestion point for potentially untrusted data that could influence agent behavior.
Ingestion points: SKILL.md (Phase 1 and Phase 2) instructs the agent to load data from sales-marketing.local.md and persona-icp to determine writing style and audience requirements.
Boundary markers: The instructions do not define delimiters or provide specific commands to the agent to treat the content of these files strictly as data, which is a common mitigation against embedded instruction following.
Capability inventory: The skill is primarily focused on text generation (articles, newsletters, scripts). The provided files do not include subprocess calls, network operations, or file-writing capabilities that could be abused through an injection.
Sanitization: The instructions do not specify any validation or sanitization steps for the data retrieved from the local context files before it is interpolated into the generation prompts.

content-creation