business-pulse-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 ingests data from external CRM datasets (e.g., rep_activity, pipeline_open) via MCP tools and synthesizes it into human-readable coaching takeaways. There are no explicit boundary markers or sanitization steps described to prevent malicious instructions embedded in CRM records from influencing the agent's behavior.
- [DATA_EXFILTRATION]: The skill is designed to retrieve sensitive business revenue and pipeline data and stage it as a Gmail draft to an external address (tkipper@epiphan.com). While this is the primary purpose for reporting, it represents a data exit point for sensitive internal information. Evidence: Delivery section in SKILL.md and Gmail tool in config.json.
- [DATA_EXFILTRATION]: The skill accesses local files outside its own directory, specifically ~/.claude/skills/epiphan-house-rules-skill/reference/identifiers.md, to retrieve mapping data for SDR seats and owner IDs, exposing internal organizational structures to the agent context. Evidence: Quick Start section in SKILL.md.
Audit Metadata