The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines blueprints for workflows (e.g., LIF-03 for reply classification and INB-01 for form enrichment) that ingest untrusted data from external sources such as email replies or form webhooks. This content is then processed by AI nodes, creating a potential surface for indirect prompt injection where malicious content in an external payload could attempt to influence the agent's logic.
Ingestion points: Webhook triggers described in references/gtm-flow-catalog.md and references/node-patterns.md (ingesting data from HubSpot, Typeform, and sequencers).
Boundary markers: The skill explicitly instructs users to implement "Normalize" steps to map data to a canonical schema and recommends the use of "Structured Output Parser" nodes to constrain AI outputs.
Capability inventory: The provided script scripts/check-output.py performs local file reads for validation. The blueprints describe workflows with extensive capabilities, including network operations via HTTP Request nodes and write access to CRM systems (HubSpot).
Sanitization: The skill emphasizes data normalization, HMAC signature verification for webhooks, and human-in-the-loop gates for high-risk actions like outbound sending.

n8n-toolkit