churn-risk
Warn
Audited by Gen Agent Trust Hub on Feb 27, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill initiates the execution of a local Python script named churn-predictor.py to process customer behavioral data. This introduces a risk of arbitrary command execution if the script is not securely managed or if its inputs are compromised.
- [DATA_EXFILTRATION]: The skill reads data from the local file system at ~/.claude-marketing/brands/ to load brand profiles and guidelines. Reading from directories outside the immediate skill folder increases the surface area for unauthorized data exposure.
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via data ingested from Salesforce or HubSpot.
- Ingestion points: CRM manager agent retrieves customer behavioral records and custom churn signals.
- Boundary markers: No markers or protective instructions (e.g., delimiters) are used to separate untrusted data from the execution prompt.
- Capability inventory: The skill can read local files and execute local scripts (churn-predictor.py).
- Sanitization: No sanitization, escaping, or validation of the external CRM data is described in the workflow.
Audit Metadata