send-notification

Pass

Audited by Gen Agent Trust Hub on Mar 5, 2026

Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill reads local application data and brand profiles from the ~/.claude-marketing/ directory to customize message delivery.
  • Data from these files is used to influence the formatting and context of notifications sent to well-known messaging platforms (Slack and Intercom).
  • [PROMPT_INJECTION]: The skill has an indirect prompt injection attack surface as it interpolates external data and user-provided messages into its execution flow.
  • Ingestion points: The skill ingests 'Notification content', 'Related context' (derived from Campaign or Approval IDs), and 'Attachments' (provided via URLs).
  • Boundary markers: The skill description does not specify the use of delimiters or 'ignore embedded instructions' markers when handling interpolated external content.
  • Capability inventory: The agent possesses capabilities to read local filesystem data (~/.claude-marketing/) and perform network operations through the Slack and Intercom MCP servers.
  • Sanitization: No explicit sanitization, validation, or escaping of ingested notification content or external context is described in the process steps.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 5, 2026, 11:19 PM