Skip to content

feat(swesmith): add SWE-Smith benchmark evaluation scaffold#428

Draft
mkaramuk wants to merge 3 commits intoOpenHands:mainfrom
mkaramuk:feat/swe-smith
Draft

feat(swesmith): add SWE-Smith benchmark evaluation scaffold#428
mkaramuk wants to merge 3 commits intoOpenHands:mainfrom
mkaramuk:feat/swe-smith

Conversation

@mkaramuk
Copy link

@mkaramuk mkaramuk commented Feb 18, 2026

Adds SWE-Smith benchmark support following the existing Evaluation base class pattern. The scaffold handles inference (running agents against SWE-Smith task instances in Docker containers) and evaluation (converting results to SWE-Smith format and invoking the swesmith harness).

Environment variable forwarding into Docker containers uses an OPENHANDS_* prefix convention — any env var with this prefix is forwarded with the prefix stripped (e.g., OPENHANDS_ANTHROPIC_API_KEY becomes ANTHROPIC_API_KEY inside the container).

Blocked on swesmith PyPI release

This PR should not be merged yet. The swesmith package on PyPI (0.0.8) does not include full private repo support, which was recently merged into SWE-smith's main branch but not yet released. Once a new version is published to PyPI, the dependency in pyproject.toml can point to it and this PR is ready to merge. That's why this PR marked as draft but we are open for any reviews, the code is ready.

reisepass and others added 3 commits February 18, 2026 13:16
Add core inference and evaluation scripts for running OpenHands agents
on SWE-Smith task instances.

Co-Authored-By: Muhammed Karamuk <mkaramuk@proton.me>
Co-Authored-By: Muhammed Karamuk <mkaramuk@proton.me>
Co-Authored-By: Rb <rubenwolff@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants