The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from a PostHog project's schema and historical records, creating an indirect prompt injection surface.
Ingestion points: The skill retrieves potentially untrusted data such as event names, property values, and previous experiment configurations using tools like experiment-setup-context, read-data-schema, experiment-saved-metrics-retrieve, and experiment-saved-metrics-list.
Boundary markers: The skill incorporates a structured decision-making process defined in references/setup-decisions.md, which requires the agent to categorize findings into "Confident", "Best guess", or "Not decided" tiers, and explicitly instructs the agent to confirm event existence before proceeding.
Capability inventory: The skill possesses the ability to write to the project state through experiment-create, experiment-update, and experiment-launch (the latter only upon explicit user request).
Sanitization: While the skill relies on the agent's interpretation of tool outputs, it enforces strict schema usage for experiment creation and provides specific logic for filtering and mapping facts to configuration settings, minimizing the impact of malicious content in processed data.

creating-experiments