The Agent Skills Directory

[SAFE]: The skill implements a 'fail-closed' policy regarding sensitive data, explicitly instructing the agent to refuse processing any raw personally identifiable information (PII) such as names, emails, or phone numbers pasted by the user.
[SAFE]: All external URLs referenced in the documentation target official government legal portals (Taiwan's moj.gov.tw) or authoritative platform support pages (Google Ads and Meta Business Help), providing verifiable and safe compliance guidance.
[SAFE]: The skill defines strict operational boundaries, prohibiting the mutation of live ad accounts and requiring that any campaign objects created through MCP tools must be set to 'paused' status by default, preventing unintended live spend or audience targeting.
[SAFE]: The provided Python snippet for character counting is a standard utility function that uses the built-in unicodedata library to handle CJK double-width character calculations for Google Ads specs; it does not introduce any risk of remote code execution or unsafe deserialization.
[SAFE]: Data minimization practices are enforced for remarketing and list uploads, including instructions to suppress opt-out lists locally rather than uploading identifying data to third-party platforms.

ads