The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted sources, which creates a potential surface for indirect prompt injection attacks. \n- Ingestion points: Data sources identified in SKILL.md and references/skill-contract.md include emails, reviews, comments, and scraped web content. \n- Boundary markers: Instructions in references/skill-contract.md define external data as evidence only and not as instructions. \n- Capability inventory: The skill provides access to external marketing and CRM platforms such as Stripe, HubSpot, and Salesforce as described in references/connectors.md. \n- Sanitization: SKILL.md contains an explicit mandate to ignore any instructions embedded within evidence. \n- [SAFE]: The skill implements security best practices by requiring explicit authorization for persistent writes and advising against storing API keys in source control.

churn-prevention