The Data Was Public. The Agent Path Wasn't. So His Mock Became My Documentation.

A developer recently contributed a pull request to an open-source project, attempting to fix issues with an AI agent's query generation. Because the agent relied on an authenticated organization-scoped endpoint that was inaccessible to the contributor, they created a local mock of the schema to test their changes. This led to a subtle but significant issue: the contributor's well-intentioned, manually defined schema contained field names that differed from the actual production database. These incorrect names were then merged into the project's tool documentation, effectively promoting assumptions to authoritative guidance. The author highlights the risks of providing reproduction paths that do not match the actual system architecture. The key takeaway for maintainers is to provide a canonical, reproducible contract or fixture for privileged dependencies, ensuring that contributors can test against the real interface without requiring internal credentials, thereby preventing the accidental propagation of incorrect assumptions into project documentation.