The Agent Skills Directory

[SAFE]: The skill uses official PostHog tools for all operations, ensuring that actions are governed by the platform's existing security and permission models.
[SAFE]: The workflow mandates several human-in-the-loop checkpoints, requiring the agent to confirm source/target projects and review configuration details with the user before performing any write operations. This effectively mitigates the risk of unauthorized or accidental data modification.
[SAFE]: New endpoints are created with materialization disabled by default, preventing unexpected resource consumption or costs in the target environment.
[PROMPT_INJECTION]: The skill processes untrusted data (endpoint names, descriptions, and queries) retrieved from the source project, which creates a surface for indirect prompt injection. This risk is addressed by the required manual review steps before the data is used to create or execute a new endpoint.
Ingestion points: endpoint-get reads query, description, name, and tags (SKILL.md).
Boundary markers: Absent; there are no instructions to the agent to treat the retrieved configuration data as untrusted content.
Capability inventory: endpoint-create and endpoint-run (SKILL.md).
Sanitization: Absent; the skill instructions indicate that the source query should be copied verbatim.

copying-endpoints-across-projects