The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill ingests data from several untrusted or external sources, including email correspondence (Gmail/Outlook), meeting transcripts/notes (Zocks), and CRM records.
Ingestion points: SKILL.md describes pulling recent correspondence and meeting summaries.
Boundary markers: The instructions do not explicitly define boundary markers or 'ignore' instructions for the content being processed by the claude-for-financial-advisors:source-extract subagents.
Capability inventory: The skill generates markdown files and can trigger conversion to .docx or .pdf via external skills.
Sanitization: While the skill uses structured schema blocks for extraction, there is no explicit instruction to sanitize or escape natural language content found in emails or notes before interpolation into the final document.
[Dynamic Subagent Execution]: The skill utilizes a parallel execution pattern using the Agent(claude-for-financial-advisors:source-extract) tool.
This is a form of dynamic delegation where specialized subagents handle the extraction of sensitive data. The skill includes good security practices by maintaining 'identity ownership' in the main agent and requiring confirmation if subagents detect identity mismatches.
[Data Privacy and Identity Verification]: The skill includes specific logic to prevent privacy incidents by requiring identity confirmation in the CRM before pulling data.
It explicitly warns that 'pulling the wrong client's data into a prep doc is a privacy incident' and mandates advisor confirmation for name matches. This demonstrates a security-conscious design for handling sensitive financial information.

pre-meeting