Security audit
Security checks for vulnerabilities and agentic risk
This skill is a coherent CRM/source-tracking integration that stores customer attribution data locally, with privacy and scoping details users should review.
Install this only in an environment where it is acceptable to persist customer identifiers, order history, source attribution, and repurchase-trigger data locally. Review the storage paths and add operational controls for retention, deletion, tenant isolation, and access to the generated CRM files.
The skill declares tools read, exec and documents persistent writes to local JSON files, yet there is no explicit permissions model or narrowing of allowed file paths/operations. This creates a real risk of unauthorized customer-data read/write behavior, especially because the skill processes tenant and customer identifiers that could influence file access decisions.
The manifest description uses broad trigger language such as post-delivery callbacks, public-account interaction, and repurchase recommendation without clearly limiting when the skill may run or what data/actions are in scope. In a CRM context handling customer records and attribution, vague triggering increases the chance of over-collection, unintended activation, and processing data outside user expectations.
The documentation explicitly states that customer source and profile data are written to local files, including per-tenant state and source-tracking mappings, but provides no user-facing notice about persistence, retention, or sensitivity. Because this is CRM data tied to customer identities and behavioral attribution, undisclosed local storage raises privacy, compliance, and accidental exposure risks.
No suspicious patterns detected.
