Propagate OpenHands MAX reasoning effort#54
Conversation
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 93b3c55e5c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| description = "OpenHands agent via ACP (multi-model, Python-based)" | ||
| install_cmd = "export DEBIAN_FRONTEND=noninteractive && export PATH=\"$HOME/.local/bin:$PATH\" && ( command -v curl >/dev/null 2>&1 && command -v git >/dev/null 2>&1 || if command -v apt-get >/dev/null 2>&1; then ( attempt=1; while [ \"$attempt\" -le 3 ]; do rm -rf /var/lib/apt/lists/*; apt-get clean; if apt-get -o Acquire::Retries=3 update -qq && apt-get -o Acquire::Retries=3 install -y -qq curl ca-certificates git; then exit 0; fi; case \"$attempt\" in 1) sleep 2; attempt=2 ;; 2) sleep 4; attempt=3 ;; *) sleep 6; attempt=4 ;; esac; done; exit 1 ); elif command -v dnf >/dev/null 2>&1; then dnf -y --allowerasing install curl ca-certificates git >/dev/null 2>&1; elif command -v apk >/dev/null 2>&1; then apk add --no-cache curl ca-certificates git >/dev/null 2>&1; else echo 'OpenHands GitHub install requires curl and git' >&2; exit 127; fi ) && ( UV_OK=0; if command -v uv >/dev/null 2>&1; then UV_VER=$(uv --version 2>/dev/null | awk '{print $2}'); if [ -n \"$UV_VER\" ] && [ \"$(printf '%s\\n' 0.11.6 \"$UV_VER\" | sort -V | head -n1)\" = \"0.11.6\" ]; then UV_OK=1; fi; fi; if [ \"$UV_OK\" = 0 ]; then curl -LsSf https://astral.sh/uv/install.sh | sh >/dev/null 2>&1 && export PATH=\"$HOME/.local/bin:$PATH\"; fi && printf 'openhands-sdk==1.22.1\\nopenhands-tools==1.22.1\\n' > /tmp/oh-sdk-overrides.txt && uv tool install --force --refresh --overrides /tmp/oh-sdk-overrides.txt --from 'git+https://github.com/OpenHands/OpenHands-CLI.git@3ca17446c5d9c1e35e054803478a3501ec251ecf' openhands --python 3.12 && uv tool list | grep -q '^openhands\\b' ) && chmod o+x /root /root/.local /root/.local/share /root/.local/share/uv /root/.local/share/uv/tools 2>/dev/null; mkdir -p ~/.openhands && echo '{\"llm\":{\"model\":\"placeholder\",\"api_key\":\"placeholder\"}}' > ~/.openhands/agent_settings.json && command -v openhands >/dev/null 2>&1" | ||
| launch_cmd = "export PATH=\"$HOME/.local/bin:$PATH\" && mkdir -p ~/.openhands && { printf '{\"llm\":{\"model\":\"%s\",\"api_key\":\"%s\"' \"$LLM_MODEL\" \"$LLM_API_KEY\"; if [ -n \"$LLM_BASE_URL\" ]; then printf ',\"base_url\":\"%s\"' \"$LLM_BASE_URL\"; fi; if [ -n \"$LLM_API_VERSION\" ]; then printf ',\"api_version\":\"%s\"' \"$LLM_API_VERSION\"; fi; printf '}}'; } > ~/.openhands/agent_settings.json && openhands acp --always-approve --override-with-envs" | ||
| launch_cmd = "export PATH=\"$HOME/.local/bin:$PATH\" && mkdir -p ~/.openhands && { printf '{\"llm\":{\"model\":\"%s\",\"api_key\":\"%s\"' \"$LLM_MODEL\" \"$LLM_API_KEY\"; if [ -n \"$LLM_BASE_URL\" ]; then printf ',\"base_url\":\"%s\"' \"$LLM_BASE_URL\"; fi; if [ -n \"$LLM_API_VERSION\" ]; then printf ',\"api_version\":\"%s\"' \"$LLM_API_VERSION\"; fi; if [ \"$LLM_REASONING_EFFORT\" = \"max\" ]; then printf ',\"litellm_extra_body\":{\"reasoning\":{\"effort\":\"max\"}}'; elif [ -n \"$LLM_REASONING_EFFORT\" ]; then printf ',\"reasoning_effort\":\"%s\",\"litellm_extra_body\":{\"reasoning_effort\":\"%s\"}' \"$LLM_REASONING_EFFORT\" \"$LLM_REASONING_EFFORT\"; fi; printf '}}'; } > ~/.openhands/agent_settings.json && openhands acp --always-approve --override-with-envs" |
There was a problem hiding this comment.
When LLM_REASONING_EFFORT=max, this branch omits the typed reasoning_effort field, but the pinned OpenHands SDK 1.22.1 defaults that field to high and forwards it on both chat and responses calls. As a result MAX runs still send a conflicting typed high effort alongside the litellm_extra_body override, so gateways/providers that honor the typed field or reject duplicate effort fields will not actually run at MAX. Explicitly clear the typed field for this branch (for example by writing it as null) before relying on the extra body.
Useful? React with 👍 / 👎.
|
Updated the branch onto current
Validation against the refreshed BenchFlow PR #921 runtime (including companion #924): manifest parity 12/12 passed. The agents registry suite is 79 passed with 2 pre-existing current-main failures ( This update was generated by Codex on behalf of Bingran You. |
|
OpenHands manifest sync update (2026-07-13): pushed
Validation:
This manifest update is the external parity dependency for This update was generated by Codex on behalf of Bingran You. |
1 participant
Summary
reasoning_effortfor values through xhighlitellm_extra_body.reasoning.effortbecause OpenHands SDK 1.22.1 rejects typedmaxValidation
git diff --checkNote: the agents repo broad contract suite currently has an unrelated existing
corust-agentacp_model_format=prefixedschema failure on main.