The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill utilizes WebFetch and WebSearch to retrieve external content based on user queries, creating an ingestion surface for untrusted data. This data is processed by an agent with powerful tools including Bash, Read, and Write. A malicious web page could attempt to influence the agent's actions through embedded instructions.
Ingestion points: Data returned by WebFetch and WebSearch tools from external websites.
Boundary markers: The skill does not explicitly define delimiters or "ignore" instructions for content retrieved via external tools.
Capability inventory: Bash, Write, Read, WebFetch, WebSearch, and various Meta Ads MCP tools (e.g., mcp__cogny__meta_ads__tool_get_pixels).
Sanitization: No explicit sanitization, validation, or filtering of external content is defined in the instructions.
[COMMAND_EXECUTION]: The skill grants access to the Bash tool. While no specific malicious commands are present in the skill body, the availability of a shell environment alongside tools that fetch untrusted web content represents a security risk surface if the agent is manipulated by external data.
[EXTERNAL_DOWNLOADS]: The skill documentation and implementation examples correctly reference official Meta API endpoints (graph.facebook.com) and developer documentation. It also links to the vendor's documentation at cogny.com. These references are necessary for the skill's primary function and target well-known, established services.

meta-capi