Practical guide · buzz and hermes
Buzz + Hermes: A Governed Capability Bridge
External hands become ungovernable when they silently take over the parent job or return prose without the declared artifact.
Why this happens
A stable capability request lets the Buzz employee remain accountable while the executor changes underneath it.
What correct behavior looks like
The Buzz owner issues a bounded request, Hermes or another provider executes it, the provider returns an artifact and receipt, and the Buzz owner continues the job. Provider failure does not become false completion.
What to do
- 1Define capability, inputs, outputs, allowed tools, side effects, timeout, and receipt fields.
- 2Keep the parent job in Buzz.
- 3Send the same contract to one provider.
- 4On evidenced failure, reroute the unchanged request.
- 5Return the artifact to the parent before any approval state changes.
How to verify it worked
- The Buzz parent remains named as owner.
- The external executor produces the declared artifact and receipt.
- Provider identity and failure state are recorded.
- Independent inspection evaluates the actual returned artifact.
Common failure states
- Provider accepts work but times out without artifact or receipt.
- Fallback changes the request and makes results incomparable.
- The orchestrator creates the content instead of returning execution to the owner.
What to do next
Reuse the bridge for another bounded job before calling the capability broadly proven.
Related guides
FAQ
Does Hermes replace the Buzz employee?
No. Hermes is an execution capability; the commissioned Buzz employee retains job ownership.
Can the provider change?
Yes, when the capability contract, evidence requirements, and authority boundaries remain stable.
FREE newsletter · Buzz Operator Brief
Practical help for people actually running Buzz agents.
Get field-tested fixes, useful workflows, practical guides, and honest notes about what we are still testing.
By subscribing, you agree to receive the Buzz Operator Brief. Unsubscribe anytime.