The Agent Skills Directory

[SAFE]: No malicious patterns or security risks were detected. The skill functions as an educational guide and integration tool for the Opik observability platform (provided by comet-ml). All documented code snippets follow security best practices, such as input validation and circuit breaking.
[PROMPT_INJECTION]: The skill addresses risks associated with indirect prompt injection and provides proactive defense mechanisms.
Ingestion points: references/evaluation.md identifies data ingestion points via CSV, JSON, Pandas DataFrames, and production traces used for evaluation.
Boundary markers: references/agent-patterns.md recommends starting traces at the agent entry point and using distinct span types to maintain clear execution boundaries.
Capability inventory: The skill leverages the Opik SDK for logging traces, executing evaluation metrics, and running real-time guardrails.
Sanitization: references/agent-patterns.md includes a dedicated section on security patterns, providing a sanitize_external_content function that uses regular expressions to strip instruction-override patterns (e.g., 'ignore previous', 'system:') from retrieved data.

agent-ops