The Agent Skills Directory

[SAFE]: The skill mandates human approval for all critical operations, explicitly stating 'Nothing changes without your yes' and 'You cannot approve on the marketer's behalf.' This provides a strong defense against autonomous misuse.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external advertising platforms (Meta, Google, TikTok) and user-provided exports. This ingestion surface is a potential vector for indirect prompt injection. However, the skill implements mitigation strategies by routing creative content through dedicated safety skills (claims-gate and ad-policy-check) and using structured templates for decision logging.
[DATA_EXFILTRATION]: All data operations involve reading from or writing to authorized platform-provided tools (meta_read, brand_knowledge_save, file_write). The skill logs performance data and decisions to the brand's coworker workspace as part of its core functionality, with no evidence of unauthorized external data transfer.
[DYNAMIC_EXECUTION]: The skill utilizes automation_upsert to generate a scheduled agent prompt. This prompt is built from a controlled template rather than arbitrary external input, minimizing the risk of dynamic code injection during automated runs.

run-paid-campaigns