The Agent Skills Directory

[SAFE]: Detailed analysis of the skill instructions and supporting files reveals no evidence of malicious behavior, such as prompt injection, unauthorized data access, or remote code execution. The skill structure is purely instructional and relies on local references and hooks to ensure output quality.
[INDIRECT_PROMPT_INJECTION]: The skill has a functional surface for indirect prompt injection as it processes user-provided inputs and optional market research. This surface is well-defended by the skill's explicit evidence discipline.
Ingestion points: User-supplied objectives and optional external research data as identified in SKILL.md and the references/skill-spec.json workflow.
Boundary markers: Mandatory classification of claims into facts, inferences, assumptions, or unknowns, providing a logical boundary for the model's processing of untrusted content.
Capability inventory: Limited to reasoning over supplied context and reading local JSON/Markdown references; no autonomous file system writes or unguided network operations are present.
Sanitization: Employs the evidence-gate and closed-loop-bias-check hooks to validate claims and detect bias before final recommendations are rendered.

commerce-media-strategy