The Agent Skills Directory

[SAFE]: The skill implements a robust human-in-the-loop safety mechanism, explicitly requiring user confirmation before executing any 'dangerous' operations that could impact financial expenditures, such as modifying budgets or creating new campaigns.
[PROMPT_INJECTION]: The skill's design includes a surface for indirect prompt injection as it is designed to ingest and analyze external data from Facebook Ads.
Ingestion points: Untrusted data enters the context via the get_ad_creatives and get_insights tools which fetch content from the Facebook Ads API.
Boundary markers: There are no specific instructions or delimiters provided to the agent to treat external ad text as untrusted or to ignore instructions embedded within creatives.
Capability inventory: The skill leverages the meta-ads MCP server, which has capabilities to modify account states, including update_adset for budget changes and pause_ad for campaign management.
Sanitization: The skill does not describe any sanitization or validation logic for the text retrieved from external ad creatives.
Note: This structural risk is a consequence of the skill's primary purpose and is effectively mitigated by the mandatory manual confirmation requirement for all sensitive actions.

ads-agent