crm-sync

Warn

Audited by Gen Agent Trust Hub on Mar 5, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local script crm-sync.py via subprocess to perform CRM status checks and deduplication tasks. This execution path handles various actions (--action get-crm-status, --action check-dedup) that are not defined within the skill's source code.
  • [DATA_EXFILTRATION]: The skill accesses several sensitive files located in the user's home directory (~/.claude-marketing/brands/), including _active-brand.json, profile.json, and _manifest.json. Accessing configuration and profile data from a hidden directory involves the exposure of brand-specific metadata and potentially sensitive context.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core functionality of ingesting external data.
  • Ingestion points: Processes data from CSV file paths and JSON arrays provided by the user or external platforms.
  • Boundary markers: No specific delimiters or instructions to ignore embedded natural language commands are defined for the data processing phase.
  • Capability inventory: Includes subprocess execution (crm-sync.py), file system read operations, and network API requests to CRM providers (Salesforce, HubSpot, Zoho, Pipedrive).
  • Sanitization: While the skill validates formats (email, phone), it lacks mechanisms to sanitize or filter out malicious natural language instructions hidden within the ingested record fields.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 5, 2026, 11:19 PM