The Agent Skills Directory

[DATA_EXPOSURE]: The skill includes a dedicated privacy section ('End of run' step 6) that explicitly prohibits the storage of personally identifiable information (PII). It instructs the agent to avoid recording names, emails, phone numbers, or private handles, and to replace sensitive names in ad titles with non-identifiable IDs.
[COMMAND_EXECUTION]: The skill utilizes platform-specific gooseworks: tools (e.g., file_list, file_read, brand_knowledge_search) for its operations. These are standard within the vendor's ecosystem for managing marketing data and workspace files.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from previous runs and external marketing sources (Meta performance data). It mitigates risks by requiring specific structured formats for 'run records' and implementing logic to verify outcomes against baseline recipes rather than relying on unverified claims.
[PERSISTENCE]: The skill implements functional persistence by saving 'run records' to the platform's brain. This is the primary intended purpose of the skill—maintaining 'memory' between sessions—and is handled through authorized platform mechanisms (save-to-brain, save-learning).

run-memory