The Agent Skills Directory

[SAFE]: The skill performs read-only diagnostic checks on a Swan workspace. It adheres to data minimization best practices by requesting small samples (page size 10 or 20) for CRM hygiene and activity logs rather than pulling full datasets.
[COMMAND_EXECUTION]: The skill uses swan-execute-code for Dimension 7 to process execution logs using the pandas library. This is a legitimate use case for analyzing truncated log data for error classification and does not involve untrusted input or shell access.
[DATA_EXFILTRATION]: While the skill accesses sensitive configuration data (CRM properties, sender status, credit balance), it does so only to generate a report for the user within the agent's context. There are no external network requests or unauthorized data transfer patterns identified.
[PROMPT_INJECTION]: No prompt injection or safety bypass patterns were detected. The instructions are focused on structured evaluation and reporting.

audit-swan