Umbrella issues for E2E

Background

We already have e2e_test.go covering previous ssh-based codeinterpreter:

Now have redesigned AgentCube, bring two scenarios:

• AgentRuntime and CodeInterpreter invocation via Agentcube Router
• Session creation and reuse via x-agentcube-session-id
• Session lifecycle and TTL semantics (at least smoke coverage)

This issue tracks adding a first-stage, minimal but concrete e2e suite for both AgentRuntime and CodeInterpreter.

Scope

AgentRuntime e2e

A1. Basic agent invocation: session creation and reuse

Precondition: AgentRuntime CR (e.g. default/echo-agent) deployed.
POST /v1/namespaces/{ns}/agent-runtimes/{name}/invocations/echo without x-agentcube-session-id.

Assert:

• HTTP 200.
• Reasonable response body (e.g. echo output).
• Response header x-agentcube-session-id is present.

Repeat call with that session header and a different request body.

Assert:

• Still HTTP 200.
• Behavior consistent with same session (e.g. same session ID; optional state check if agent supports state).

A2. Missing/invalid AgentRuntime

• Call POST on a non-existent {name}.
• Assert:
  • HTTP 404 (or the documented error code).
  • Clear error payload (e.g. “AgentRuntime not found”).

A3. Idle session / TTL behavior (smoke test)

• Create a session as in A1 and capture SESSION_ID.
• Wait long enough for idle TTL + cleanup.
• Call with the same SESSION_ID.
  -Assert:
  • Either error indicating session not found.

CodeInterpreter e2e

B1. Simple code execution with session auto-creation

Precondition: CodeInterpreter CR (e.g. default/python-interpreter) deployed.
POST /v1/namespaces/{ns}/code-interpreters/{name}/invocations/run without x-agentcube-session-id.
Body: simple code snippet (e.g. print(1+1)).

• Assert:
  • HTTP 200.
  • Output (e.g. "stdout":"2\n").
  • Response header x-agentcube-session-id is present.

B2. Stateful interpreter session (multi-step)

• Reuse the SESSION_ID from B1.
• Second call: code = "x = 10".
• Third call: code = "print(x)".
• Assert:
  • Third response prints 10 (or explicitly assert stateless behavior if we decide CodeInterpreter is stateless initially).

B3. File-based workflow via CodeInterpreter (Fibonacci JSON)

• Use CodeInterpreter API to:
• Upload a small Python script (fibonacci.py) into /workspace.
• Execute it to generate /workspace/output.json.
• Retrieve the JSON result.
• Assert:
  • retrieve the result to check.

Corner cases:

1. longer delay task execution
2. warmer pool/non warmpool