The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the vendor-provided cx CLI toolset (specifically cx ai-center and cx spans) to manage configurations and query telemetry data. These operations are scoped to the Coralogix platform and follow established security practices, including a requirement for interactive confirmation on destructive actions unless explicitly overridden by the user.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze AI conversation transcripts from telemetry spans, which constitutes a vulnerability surface for untrusted external data.
Ingestion points: Conversation messages are read from gen_ai.input.messages and gen_ai.output.messages tags within GenAI spans as documented in SKILL.md and references/ai-center-queries.md.
Boundary markers: The provided DataPrime query library includes specific filters and logic (e.g., role-based filtering, text-part extraction) to isolate user and assistant turns, helping to mitigate the risk of processing malicious control instructions embedded in the data.
Capability inventory: The agent has the ability to run telemetry queries and configure AI Center policies. Security is maintained by instructions that require the agent to sample data rather than performing exhaustive scans of untrusted content.
Sanitization: The skill documentation provides technical guidance on removing system prompts and tool traffic from the analysis context to prevent instruction confusion.

cx-ai-center