Manifestwatch

Teams increasingly depend on public agent-tool servers whose names, parameter schemas, transport and error behavior can change without a lockfile or stable version contract. Manifestwatch proposes a monitor that snapshots authorized public manifests, normalizes them, produces semantic diffs and maps changed tools back to the team's own declared agent and prompt dependencies. The supplied research found no dedicated manifest-drift monitor and confirms several recent protocol and endpoint changes.

A manifest difference is not proof that production broke. Endpoint observation, fetched document, normalized snapshot, semantic diff, breaking-change candidate, dependency match, replay result, owner review, alert delivery, acknowledgement, remediation and production recovery must remain separate. Public availability does not authorize abusive polling, and a declared schema does not reveal all runtime behavior.

The technology is copyable and one referenced capability remains unverified. The compounding asset is a historical archive and dependency graph, but its value depends on faithful timestamps, fetch evidence and coverage. Numeric competitive prices are omitted because they are observed market references, not fixed product pricing.

An engineering or agent-platform team that depends on several public tool servers and needs advance evidence of contract drift across its own workflows.