The Agent Skills Directory

[SAFE]: The skill consists of pedagogical content designed to teach marketing teams about AI capabilities, limitations, and safe use. It contains no instructions to bypass safety filters, no obfuscated content, and no unauthorized network or file system operations.
[NO_CODE]: The skill is composed entirely of markdown documentation and instructional references. There are no Python scripts, Node.js packages, or shell commands associated with the skill's operation.
[INDIRECT_PROMPT_INJECTION]: The skill features a surface for processing user-supplied data (such as client names and marketing goals) to generate a customized training guide.
Ingestion points: User input is collected via the Intake questions in references/training-guide-plan.md and the Required Inputs table in SKILL.md.
Boundary markers: The skill explicitly instructs users to "stop before exercises put real client personal data into public AI tools" and to use synthetic data instead. It also references Uganda DPPA 2019 compliance.
Capability inventory: The skill's capabilities are limited to text generation for facilitator guides; it has no access to sensitive system tools, network exfiltration capabilities, or persistent storage.
Sanitization: The workflow mandates a "Human Quality Standard" review and an "Approval Rule" ensuring no content is published without full human oversight.

training-ai-foundations