cendor-sdk 1.20.0

An agent can have an identity, not just a name — and every governance row says which agent it was.

Agent(id=…)

A name is a label: two agents in two apps can share one, and renaming an agent loses its history. Pass
id= and it is emitted as gen_ai.agent.id on every span of that agent's turns — the live tree, the
post-hoc span_tree, and the governance rows. Use your own registry key, or the id a framework already
owns (Azure AI Foundry's agent_id, Bedrock's agentId, an OpenAI assistant_id).

When you give no id the attribute is simply omitted — never a hash of the name, never a
placeholder. Cendor does not invent identity.

id is appended at the end of the Agent field list, so Agent("support", "gpt-4o", "You are…")
keeps working unchanged.

The actor on every governance row (needs cendor-core 1.14 / cendor-acttrace 1.13)

The SDK's ambient provider now stamps the acting agent's id beside its name, which is what lets a
governance.* span or an audit.* mirror entry name the agent — including on the entry types that
carry no agent field of their own, such as a budget block. Measured before this release: 13 of 386
governance rows named their agent.

Also

Requires cendor-core >= 1.14 and cendor-acttrace >= 1.13. Note the behaviour change in
cendor-core 1.14.0: core.trace() now opens a real parent span, so a libs-side trace() scope
groups its calls into one trace. It opens no span inside an SDK run — your run already owns its trace.