User request
Add E2E to CI in llm-proxy. It should trigger tests that do end-to-end agent execution starting from a thread message and waiting for the response.
Specification (research)
- llm-proxy already has an E2E GitHub Actions workflow (
.github/workflows/e2e.yml) that checks out agynio/e2e and runs devspace run ci-e2e.
- The requested ‘thread message → agent executes → wait for response’ behavior spans multiple services (threads, agents, agents-orchestrator, runner, llm, llm-proxy), so the deterministic E2E test should live in agynio/e2e (go-core suite) and be invoked from llm-proxy’s E2E workflow.
Test scenario (minimal + deterministic)
Create:
- user identity
- API token
- organization
- LLM provider using deterministic endpoint
https://testllm.dev/v1/org/agynio/suite/agn/responses and protocol RESPONSES
- model with remote name
simple-hello
- agent using that model, with env
LLM_API_TOKEN=<created api token>
- thread with participants [user, agent]
Then:
- send a user thread message with body
hi
- poll
ThreadsService.GetMessages until a message exists with:
sender_id == agentID
created_at >= sentUserMessage.created_at
- body exactly
Hi! How are you?
Cleanup:
- archive thread
- delete agent
- cleanup workloads matching agent/thread labels
Timeouts:
- overall test ~6m, poll interval 2s, max wait 5m.
CI wiring
- Update llm-proxy
.github/workflows/e2e.yml to run both:
- existing
TestLLMProxyGatewayCreatedModel
- new
TestAgentRespondsToThreadMessageViaLLMProxy
- Implement via
E2E_GO_TEST_RUN='TestLLMProxyGatewayCreatedModel|TestAgentRespondsToThreadMessageViaLLMProxy' (or equivalent list supported by the harness).
Acceptance criteria
- Workflow triggers and runs the new test.
- Test validates the full message→agent→response loop deterministically without commercial LLM credentials.
- CI artifacts/diagnostics are uploaded on failure (existing behavior).
User request
Add E2E to CI in llm-proxy. It should trigger tests that do end-to-end agent execution starting from a thread message and waiting for the response.
Specification (research)
.github/workflows/e2e.yml) that checks out agynio/e2e and runsdevspace run ci-e2e.Test scenario (minimal + deterministic)
Create:
https://testllm.dev/v1/org/agynio/suite/agn/responsesand protocolRESPONSESsimple-helloLLM_API_TOKEN=<created api token>Then:
hiThreadsService.GetMessagesuntil a message exists with:sender_id == agentIDcreated_at >= sentUserMessage.created_atHi! How are you?Cleanup:
Timeouts:
CI wiring
.github/workflows/e2e.ymlto run both:TestLLMProxyGatewayCreatedModelTestAgentRespondsToThreadMessageViaLLMProxyE2E_GO_TEST_RUN='TestLLMProxyGatewayCreatedModel|TestAgentRespondsToThreadMessageViaLLMProxy'(or equivalent list supported by the harness).Acceptance criteria