The Agent Skills Directory

[SAFE]: The skill is authored by the vendor (PostHog) and interacts exclusively with the platform's API via MCP tools (experiment-create, read-data-schema, etc.). It adheres to the primary purpose of experiment management without attempting to access sensitive local files or execute arbitrary shell commands.
[INDIRECT_PROMPT_INJECTION]: The skill implements logic to ingest and interpret external project data (e.g., event schemas and previous experiment results) to automate configuration decisions. While this technically creates a surface for indirect prompt injection if a PostHog project contained maliciously named events or metrics, the risk is negligible as the skill uses this data for structural JSON payload construction rather than as direct instructions.
Ingestion points: Data returned from experiment-setup-context, read-data-schema, and experiment-saved-metrics-list (SKILL.md, references/setup-decisions.md).
Boundary markers: None explicitly defined for tool outputs; the agent is expected to parse the data according to the provided reference logic.
Capability inventory: Capabilities are limited to standard PostHog API operations such as experiment-create and experiment-update.
Sanitization: The skill relies on structured data interpretation and provides specific logic (mapping facts to tiers like 'Confident' or 'Best guess') to handle potentially ambiguous or missing data points.

creating-experiments