The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of workflows that ingest untrusted external data (event properties and person profiles) and execute side effects like email, SMS, and webhooks.
Ingestion points: Workflow triggers defined in SKILL.md and references/graph-schema.md process live event streams and person property bags (event, person, groups).
Boundary markers: The skill instructs the use of specific delimiters for templating, such as {person.property} (HogQL) and {{person.property}} (Liquid), which helps the underlying engine distinguish between data and instructions.
Capability inventory: Workflows created via these instructions have the ability to send emails (function_email), SMS (function_sms), push notifications (function_push), and call external webhooks, as detailed across all referenced files.
Sanitization: The skill explicitly states that the PostHog server owns and manages bytecode compilation and performs strict schema validation for every node in the graph, preventing the injection of malicious execution logic into the workflow definition itself.

building-workflows