The Agent Skills Directory

[SAFE]: The skill is composed of static Markdown instructions and reference templates. It does not include any executable code (e.g., Python, JavaScript, or Shell scripts), significantly reducing the risk of runtime attacks such as remote code execution or privilege escalation.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves processing external data like social media mentions, customer reviews, and user-generated content (UGC). While this introduces a surface for indirect prompt injection, the risk is mitigated by the skill's requirement to follow global interaction and safety protocols (_system/interaction-protocol.md). The processing of this data is essential to the skill's purpose, and no exploitable capabilities were found.
[EXTERNAL_DOWNLOADS]: References to external services (e.g., Shopify, Amazon, Meta, and reputation monitoring tools) are limited to strategic recommendations for the user or links to official support and reporting pages. No unauthorized or suspicious external downloads are initiated by the skill.
[DATA_EXFILTRATION]: There is no evidence of attempts to harvest or exfiltrate sensitive information. The skill manages logs and learnings locally within its designated directory structure (brand-brain/pr/) and explicitly prohibits the use of unauthorized private contact information.

afa-pr