The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill manages workflows that ingest and process external data (event payloads, person properties, and session recordings) which are then used in logic branches or interpolated into message templates.
Ingestion points: External data enters the agent context via trigger event payloads (event), person properties (person), and session recordings analyzed by the Replay Vision step (specified in SKILL.md and references/graph-schema.md).
Boundary markers: There are no explicit instructions for the agent to wrap external content in delimiters or include warnings to ignore embedded instructions in the generated workflow configurations.
Capability inventory: The skill utilizes tools like workflows-create, workflows-patch-graph, workflows-patch-action-email, workflows-enable, and workflows-run-batch to modify production workflows and send real-world messages (email, SMS, push notifications) as documented in references/lifecycle-and-debugging.md.
Sanitization: The instructions do not mention sanitization, escaping, or filtering of external content before it is interpolated into templates or used for AI-based analysis in Replay Vision.
[SAFE]: The skill is authored by a recognized vendor (posthog) and all referenced tools and resources are consistent with the vendor's platform functionality. The workflows are managed via a structured JSON graph format and use internal templating engines (HogQL, Liquid) managed server-side, which is standard for the application's primary purpose.

building-workflows