personalization-at-scale-skill
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from prospect lists and LinkedIn profiles to generate personalized outreach lines.
- Ingestion points: The skill ingests data from uploaded CSV files and researches live LinkedIn activity, posts, and comments (SKILL.md).
- Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish between its primary instructions and potentially malicious instructions embedded in the external research data.
- Capability inventory: The skill utilizes tool calls (
mcp__claude_ai_Epiphan_Ai__hubspot_search_contacts) to search CRM data and is instructed to write analytics to the local file system. - Sanitization: No sanitization or filtering of the retrieved LinkedIn content or CSV data is defined before the agent processes it to generate text.
- [SAFE]: The skill includes instructions to log performance metrics to a local analytics file at
~/.claude/skill-analytics/last-outcome-personalization-at-scale.json. This is a standard telemetry pattern for tracking skill efficacy and does not involve network exfiltration or sensitive credential harvesting.
Audit Metadata