i do not think an agent should receive broad publishing access by default.
publishing changes the package supply chain.
it deserves an explicit path.
that does not mean agents cannot help with releases.
it means the permissions should match the work.
with omni line, i use personal access tokens for package manager traffic.
i can give an agent registry:read when it needs to install internal dependencies, inspect a package, or prepare a change.
i only use registry:write for a workflow that is intended to publish.
i keep that publishing token separate from ordinary read access.
the token itself is not the only safeguard.
omni line intersects its scopes with the user’s role in the organization.
that keeps an agent from gaining more registry permission than the underlying role permits.
i also make the task explicit.
i can ask an agent to update a version, run tests, build the package, and prepare a release.
i can ask it to publish only when publishing is the requested action.
i do not make publish an implied final step.
this matters when the project has automation across developer machines, ci runners, and agents.
they can all use the same npm registry URL.
they can all use familiar npm commands.
their tokens can still represent different responsibilities.
i can revoke a dedicated workflow token when the automation changes.
i can create a new one with a narrower scope when the task changes.
i can keep the complete token out of the repository because it is only shown once at creation.
that is the publishing model i want for agents.
the agent can do real release work.
the team can still decide what it is allowed to publish and when.
automation gets a path forward.
human ownership of package access remains clear.
Top comments (0)