The Agent Skills Directory

[SAFE]: The skill focuses on operational control and reconciliation of evaluation data using established project mechanisms, with no executable code or dangerous instructions.
[DATA_EXFILTRATION]: The instructions explicitly advise keeping credentials in secret mechanisms, redacting sensitive information (personal identifiers, signed URLs) from logs, and using only approved storage and destinations.
[INDIRECT_PROMPT_INJECTION]: While the skill processes potentially untrusted evaluation outputs and logs, it includes guidelines for reconciliation, evidence-based reporting, and deterministic checks, which mitigates the risk of acting on untrusted data.
[PROMPT_INJECTION]: No attempts to override agent behavior, bypass safety guidelines, or extract system prompts were detected in the instructions or metadata.

evaluation-operations