meddic-call-prep-auto

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONPERSISTENCE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from multiple untrusted external sources, which may contain malicious instructions that could manipulate the agent's output or behavior.
  • Ingestion points: Data is gathered from gcal_get_event (descriptions), hubspot_get_deal (notes), clari_get_call_summary, and search_threads (Gmail content).
  • Boundary markers: The instructions do not specify any delimiters or instructions for the agent to ignore potentially malicious content within the retrieved data.
  • Capability inventory: The skill has broad read access to sensitive business tools (Gmail, Calendar, HubSpot, Clari, Apollo) and performs local file writing for reporting.
  • Sanitization: There is no evidence of sanitization, filtering, or validation of the data retrieved from external APIs before it is synthesized into the MEDDIC brief.
  • [DATA_EXFILTRATION]: The skill accesses and aggregates high-sensitivity information across multiple disjointed business platforms.
  • Evidence: The workflow retrieves and combines content from private emails, calendar event details (attendees and descriptions), and detailed CRM activity history. While this facilitates the 'Call Prep' goal, it creates a large sensitive data footprint within a single prompt context.
  • [PERSISTENCE]: The skill implements a mechanism to write data to the local filesystem across different sessions.
  • Evidence: The 'Emit Outcome Sidecar' step explicitly instructs the agent to write a JSON file containing execution metrics and status to ~/.claude/skill-analytics/last-outcome-meddic-call-prep-auto.json.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 09:56 AM
Security Audit — agent-trust-hub — meddic-call-prep-auto