Add integration-style tests for conversation restore behavior

[enyst]

HUMAN: This PR proposes a follow-up to SystemPromptEvent immutability issue, previous removal of persisted enforcement, and conversation restore:

• cross-package tests for the behaviors we want on conversation restore
• to lock in invariants.

---

Summary

Adds an integration-like test module that documents the expected behavior of LocalConversation restore.

Covered behaviors

• Happy path lifecycle: start -> send/run -> send/run -> close -> restore -> send/run
• Restore fails when the runtime agent toolset changes (removed/added tools)
• Restore succeeds when non-breaking agent config changes (LLM model, condenser max_size, AgentContext/skills)
• and others.

Notes

• Uses patched litellm_completion and create_mock_litellm_response for deterministic, offline runs.
• Includes a small RestoreLifecycle helper to keep the behavior spec readable.

@enyst can click here to continue refining the PR

---

Agent Server images for this PR

• GHCR package: https://github.com/OpenHands/agent-sdk/pkgs/container/agent-server

Variants & Base Images

Variant	Architectures	Base Image	Docs / Tags
java	amd64, arm64	eclipse-temurin:17-jdk	Link
python	amd64, arm64	nikolaik/python-nodejs:python3.12-nodejs22	Link
golang	amd64, arm64	golang:1.21-bookworm	Link

Pull (multi-arch manifest)

# Each variant is a multi-arch manifest supporting both amd64 and arm64
docker pull ghcr.io/openhands/agent-server:20bdc38-python

Run

docker run -it --rm \
  -p 8000:8000 \
  --name agent-server-20bdc38-python \
  ghcr.io/openhands/agent-server:20bdc38-python

All tags pushed for this build

ghcr.io/openhands/agent-server:20bdc38-golang-amd64
ghcr.io/openhands/agent-server:20bdc38-golang_tag_1.21-bookworm-amd64
ghcr.io/openhands/agent-server:20bdc38-golang-arm64
ghcr.io/openhands/agent-server:20bdc38-golang_tag_1.21-bookworm-arm64
ghcr.io/openhands/agent-server:20bdc38-java-amd64
ghcr.io/openhands/agent-server:20bdc38-eclipse-temurin_tag_17-jdk-amd64
ghcr.io/openhands/agent-server:20bdc38-java-arm64
ghcr.io/openhands/agent-server:20bdc38-eclipse-temurin_tag_17-jdk-arm64
ghcr.io/openhands/agent-server:20bdc38-python-amd64
ghcr.io/openhands/agent-server:20bdc38-nikolaik_s_python-nodejs_tag_python3.12-nodejs22-amd64
ghcr.io/openhands/agent-server:20bdc38-python-arm64
ghcr.io/openhands/agent-server:20bdc38-nikolaik_s_python-nodejs_tag_python3.12-nodejs22-arm64
ghcr.io/openhands/agent-server:20bdc38-golang
ghcr.io/openhands/agent-server:20bdc38-java
ghcr.io/openhands/agent-server:20bdc38-python

About Multi-Architecture Support

• Each variant tag (e.g., 20bdc38-python) is a multi-arch manifest supporting both amd64 and arm64
• Docker automatically pulls the correct architecture for your platform
• Individual architecture tags (e.g., 20bdc38-python-amd64) are also available if needed

[all-hands-bot]

Good test coverage for conversation restore behavior! The tests clearly document the expected semantics. Found a few minor issues to address.

[enyst]

@OpenHands Install yourself gh if you don't have it, then check this PR, understand it deeply, and pull the AI reviewer comments, also inline review threads comments.

Keep in mind it's AI, so it can be wrong and we need to verify each comment. If valid, let's fix them (and mark it solved), otherwise just reply with your opinion to the comment.

[openhands-ai]

I'm on it! enyst can track my progress at all-hands.dev

[enyst]

Resolved AI review comment r2720748938 by hoisting create_mock_litellm_response import to module scope (commit 7ed20a1). For r2720748971: keeping getattr() is intentional because event type/shape may vary; we only assert the two relevant fields when present.

[openhands-ai]

Final summary (new work since the last summary)

Verification

• Fetched AI reviewer inline threads via GitHub API/GraphQL and re-validated each comment.
• Applied valid comments as minimal, focused code changes (no behavior changes beyond test cleanup/clarity).
• Kept changes concise: only edited tests/cross/test_conversation_restore_behavior.py.
• Ran the test module locally: pytest tests/cross/test_conversation_restore_behavior.py → 8 passed.
• Pushed updates to the existing PR branch test/conversation-restore-behavior.
• Marked review threads as resolved (including the remaining one after replying).

Code changes made (since last summary)

All included in commit d76eba8f:

• Removed dead code in multiple pytest.raises(...) blocks (restored.close() after an exception) by calling lifecycle.restore(...) directly.
• Added a short comment explicitly documenting tool-order independence in the happy-path test (persisted vs runtime tool list order).
• Simplified a minor assertion: replaced pytest.approx(0.42) with == 0.42.

Notes on PR interactions

Replied on the remaining “getattr vs explicit event type” thread to explain why keeping getattr is intentional (resilience to event model refactors while still asserting the relevant fields), then resolved it.

No further diffs remain on the branch (working tree clean, branch already pushed).

View full conversation

[all-hands-bot]

Overall this is a well-structured test suite that clearly documents conversation restore behavior. Found a few issues that should be addressed.

[raymyers]

Neat approach!