v2.4.6

What's Changed

• fix(agents): count billed calls that fail to parse against cost_limit by @EnesYilmazcode in #915

Full Changelog: v2.4.5...v2.4.6

v2.4.5

What's Changed

• fix(config): read config files as UTF-8 regardless of locale by @EnesYilmazcode in #892
• fix(models): don't mask malformed tool-call errors as truncation by @klieret in #896

Full Changelog: v2.4.4...v2.4.5

New contributors

Thanks to simple-zy for raising issue #894

v2.4.4

What's Changed

• fix: remove startup-banner emoji that crashes import on non-UTF stdout by @EnesYilmazcode in #886
• fix(benchmarks): don't mutate shared config across worker threads in get_sb_environment by @EnesYilmazcode in #887

New Contributors

• @EnesYilmazcode made their first contribution in #886

Full Changelog: v2.4.3...v2.4.4

v2.4.3

What's Changed

Fixes

• fix(run): pass the startup_command as a dict to env.execute() function. by @Ethan-Kuang-6666 in #880

Enhancements

• Reject empty config override keys by @CodingFeng101 in #883

New Contributors

• @Ethan-Kuang-6666 made their first contribution in #880
• @CodingFeng101 made their first contribution in #883

Full Changelog: v2.4.2...v2.4.3

v2.4.2

What's Changed

Small release to bump image versions in programbench runner.

• Update: programbench uses v6 images by @klieret in #867

Full Changelog: v2.4.1...v2.4.2

v2.4.1

What's Changed

• fix: kill local process group on timeout by @Swing-G in #865

New Contributors

• @Swing-G made their first contribution in #865
• maaz139 for reporting the timeout bug

Full Changelog: v2.4.0...v2.4.1

v2.4.0

What's Changed

• Change(swe-bench configs): Drop T=0 default in swe-bench by @klieret in #861
• Enh: Expose finish_reason to format_error_template by @klieret in #862
• feat(inspector): display reasoning_content with toggle and hot-reload by @AlexanderDefuria in #815
• Enh: Cap consecutive format errors by @klieret in #863

New Contributors

• @AlexanderDefuria made their first contribution in #815

Full Changelog: v2.3.1...v2.4.0

v2.3.1

What's Changed

• Fix(swebench): source ~/.bashrc via BASH_ENV to activate testbed env by @seek-far in #858
• fix: persist full model response in extra dict on FormatError (#784) by @VittoriaLanzo in #821

New Contributors

• @VittoriaLanzo made their first contribution in #821
• @Entrpi made their first contribution in #850
• @seek-far made their first contribution in #858

Full Changelog: v2.3.0...v2.3.1

Run mini-swe-agent on ProgramBench!

What's Changed

The main feature is compatibility with ProgramBench, a new and ultra-challenging software benchmark.

Fixes

• fix: add exist_ok=True to mkdir in BubblewrapEnvironment by @hobostay in #802
• Fix/cost limit zero by @klieret in #825
• fix: add wall-clock time limit to properly kill agents by @klieret in #832

Full Changelog: v2.2.8...v2.3.0

v2.2.8

Warning

The dependency litellm in versions 1.82.7 and 1.82.8 was compromised. If you updated/installed litellm in the last few overs
SSH and API keys might have been compromised.

The compromised versions should have been removed from pypi, but the PR below also excludes them in the dependencies.

• security: exclude compromised litellm versions 1.82.7 and 1.82.8 by @michalcichon in #794

New Contributors

• @michalcichon made their first contribution in #794

Full Changelog: v2.2.7...v2.2.8