The Agent Skills Directory

Indirect Prompt Injection: The skill processes data from diverse external sources including CRM records, ad platforms, email inquiries, and public reviews. Instructions maliciously hidden in these sources could potentially influence the agent's behavior during analysis.
Ingestion points: Data is ingested from HubSpot, Mailchimp, Gmail, Microsoft 365, TikTok Ads, and public web reviews as described in SKILL.md and reference/data_sources.md. It also supports manual file uploads of sales and deal reports.
Boundary markers: No specific delimiters or 'ignore instructions' directives are specified for the data ingestion process in the provided documentation.
Capability inventory: The skill uses WebFetch for network retrieval, Read for file access, and can perform outward writes to platforms like Slack, Notion, and Canva.
Sanitization: The skill lacks explicit sanitization or filtering protocols for external content.
External Data Connectivity: The skill utilizes several third-party connectors (such as HubSpot, QuickBooks, and Shopify) to retrieve business-sensitive metrics. While central to its purpose, this involves interaction with multiple external service providers and the aggregation of potentially sensitive business information.
Network Operations: The skill uses WebFetch to gather public review data and interacts with various APIs for marketing data. Outward data transmission to Slack is protected by an explicit user confirmation gate as described in Step 7 of the main instructions.

growth-pulse