meddic-call-prep-auto-skill
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from Google Calendar descriptions, HubSpot deal notes, Apollo enrichment data, and Clari call summaries, which are interpolated into business briefs without explicit boundary markers or sanitization of the input data.
- Ingestion points: Found in
gcal_get_event,hubspot_search_deals,apollo_organizations_enrich, andclari_get_call_summarytools as described in Stage 1 of the workflow. - Boundary markers: Absent. The MEDDIC Synthesis (Stage 2) and Output Format (Stage 3) directly interpolate external strings into the prompt template.
- Capability inventory: Includes
create_draft(Gmail MCP) andask_agent(CRM data warehouse queries) in Stage 4. - Sanitization: The
check_my_copytool is mentioned for prospect-facing snippets but does not cover the primary synthesis of the brief. - [DATA_EXFILTRATION]: The skill is configured to automatically stage Gmail drafts containing sensitive prospect and deal intelligence to a hardcoded email address (
tkipper@epiphan.com). - [DATA_EXPOSURE]: The skill records execution outcomes and processed metrics into local JSON files at
~/.claude/skill-runs/and~/.claude/skill-analytics/, which may contain summaries of deal context.
Audit Metadata