The Agent Skills Directory

[DATA_EXFILTRATION]: The skill transmits target addresses and context strings to the third-party Globalping service. It includes a specific "Public targets only" section that instructs the agent to validate and refuse internal addresses (RFC1918, loopback, etc.) locally before transmission to ensure privacy and prevent disclosure of internal naming.
[CREDENTIALS_UNSAFE]: The skill manages a GLOBALPING_TOKEN environment variable for authentication. It proactively warns that the limits tool output contains a fragment of the token and directs the agent to prevent this data from being included in user-facing output, tickets, or public channels.
[PROMPT_INJECTION]: The skill processes data from external measurement probes, representing a surface for indirect prompt injection.
Ingestion points: External measurement results returned from the Globalping API via the globalping-mcp tool (SKILL.md).
Boundary markers: Not explicitly defined for tool output processing.
Capability inventory: Capabilities are limited to network diagnostics (ping, traceroute, DNS, MTR, HTTP) via MCP tool calls; no filesystem write or arbitrary code execution capabilities are granted.
Sanitization: Includes explicit instructions to sanitize the context field of business-sensitive details and filter target inputs locally before they reach the external service.

globalping-external-checks