prospect-refresh-skill
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external data (prospect names, titles, and company info) from the Apollo API and incorporates it into generated email drafts. This represents a potential indirect prompt injection surface where malicious content in an external profile could influence agent behavior.\n- Ingestion points: Apollo Mixed People API search results (Stage 2 in SKILL.md).\n- Boundary markers: The skill uses bracketed placeholders (e.g., [FIRST_NAME], [UNIVERSITY_NAME]) within skeletons in reference/email-skeletons.md to delimit where external data is inserted.\n- Capability inventory: The skill possesses the ability to create Gmail drafts (create_draft), execute logic for lead qualification (qualify_lead), and write structured state/report files to the local file system (~/.claude/skill-state/).\n- Sanitization: The skill implements a mandatory Gate B using the check_my_copy tool to review every generated email body for brand compliance and safety before creating the draft. Additionally, the drafts are not sent automatically, requiring manual user review.
Audit Metadata