The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Automated security scans identified the external resource https://pro.formview.io/assets/email.html, which is referenced as a default template in references/action-types.md, as a phishing threat. Users and agents should be cautious when accessing or configuring this endpoint.
[COMMAND_EXECUTION]: The file references/action-types.md was flagged by reputation scanners as containing malicious content. This file provides detailed instructions for configuring high-privilege operations, such as the transform feature for executing JavaScript strings in an isolated VM and the sqlconnector action for executing direct SQL queries against external databases.
[DATA_EXFILTRATION]: The skill describes how to configure webhook and email actions that can transmit form submission data to external URLs. While documented as core platform functionality, the instructions lack validation steps to ensure that data is only sent to authorized and secure endpoints, creating a potential for data exfiltration.
[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by processing external submission data through multiple active components.
Ingestion points: Submission data retrieved via the action_get and action_list tools.
Boundary markers: Absent. The skill does not provide delimiters or instructions for the agent to ignore or sanitize embedded commands within submission content.
Capability inventory: The skill allows the configuration of action_create and action_update to deploy email, webhook, sqlconnector, and transform logic.
Sanitization: There is no documented mechanism or instruction for escaping user-supplied data before it is interpolated into email templates ({{ data.fieldKey }}) or webhook payloads.

formio-actions