Boundary
Lifecycle hooks use runtime-scoped credentials.
FRAMEWORK
OpenAI Agents
Attach exact behavior context at agent lifecycle boundaries without turning Vira into the agent runtime or tool orchestrator.
EXECUTION BOUNDARY
Vira resolves at the selected agent lifecycle boundary and records the resulting execution afterward. Agent planning, tools, handoffs, and runner semantics remain in the agent framework.
IDENTITY PRESERVED
Keep the behavior identity attached to the host framework.
- agent-start resolve identity
- Behavior Point
- exact behavior revision
- trace identity
SUPPORTED OPERATIONS
Use Vira where the control boundary actually belongs.
- 01
resolve at agent start
- 02
carry returned context through the run
- 03
trace the agent result at agent end
- 04
preserve exact behavior identity for later review
IMPLEMENTATION HOOK
Keep the host runtime native.
This example mirrors the repository documentation. The exact API contract remains in Docs and source.
const hooks = createOpenAIAgentsHooks(vira, {
context: { kind:"agent", host:{ name:"openai-agents", runtime:"node" }, modelTarget:{ provider:"openai", model:"your-model-id" } },
});
const context = await hooks.onAgentStart(resolveRequest);
const output = await runner.run(agent, input);
await hooks.onAgentEnd(context, output);SECURITY BOUNDARY
Integration does not equal authority.
Boundary
Publishing and high-impact control operations stay outside the agent runner.
LIMITATIONS
State what the adapter does not control.
- Vira does not become the agent planner.
- Fail-open metadata-only traces cannot become authoritative Teach evidence.
- Agent tool authority still requires explicit registry/control semantics.