microsoft/debug-gym 1.3.0

1.3.0

Repository: microsoft/debug-gym

Tag: 1.3.0

Published: 2026-01-09T16:52:06Z

Prerelease: no

Release notes:

What's Changed

• Specify base images for mini-nightmare and aider by @chisingh in https://github.com/microsoft/debug-gym/pull/255
• Support env variable in pod_spec_kwargs by @MarcCote in https://github.com/microsoft/debug-gym/pull/248
• HuggingFace LLM by @xingdi-eric-yuan in https://github.com/microsoft/debug-gym/pull/251
• Fverac/swebench self test agent by @fverac in https://github.com/microsoft/debug-gym/pull/253
• make eval tool parameterizable by @fverac in https://github.com/microsoft/debug-gym/pull/269
• Clean space in GA runner by @MarcCote in https://github.com/microsoft/debug-gym/pull/270
• update trajectory vis by @xingdi-eric-yuan in https://github.com/microsoft/debug-gym/pull/278
• Support loading local parquet dataset by @MarcCote in https://github.com/microsoft/debug-gym/pull/272
• Fix Kube double reservation by @sordonia in https://github.com/microsoft/debug-gym/pull/285
• Use None is info doesn't exists in base agent by @MarcCote in https://github.com/microsoft/debug-gym/pull/290
• Fixing things for tutorial demo by @MarcCote in https://github.com/microsoft/debug-gym/pull/289
• Agent args dataclass by @sordonia in https://github.com/microsoft/debug-gym/pull/287
• Envs for images by @sordonia in https://github.com/microsoft/debug-gym/pull/291
• Simpler agent by @sordonia in https://github.com/microsoft/debug-gym/pull/294
• Deal with negative history cutoff + other fixes by @MarcCote in https://github.com/microsoft/debug-gym/pull/303
• Make sure we reset the tests before running the evaluation by @MarcCote in https://github.com/microsoft/debug-gym/pull/309
• Override setLevel to only affect console output, preserving file logging by @Copilot in https://github.com/microsoft/debug-gym/pull/301
• Add script to replay a trajectory. by @MarcCote in https://github.com/microsoft/debug-gym/pull/279
• Add unit tests for Bash file creation with EOF and newline handling by @Copilot in https://github.com/microsoft/debug-gym/pull/296
• allow rewrite to create new files by @fverac in https://github.com/microsoft/debug-gym/pull/240
• Better json viewer by @xingdi-eric-yuan in https://github.com/microsoft/debug-gym/pull/308
• Free Env by @xingdi-eric-yuan in https://github.com/microsoft/debug-gym/pull/284
• Error Handling by @xingdi-eric-yuan in https://github.com/microsoft/debug-gym/pull/311
• Fix AnthropicLLM extended thinking support and improve error retry logic by @xingdi-eric-yuan in https://github.com/microsoft/debug-gym/pull/316
• step / init / execute_action by @sordonia in https://github.com/microsoft/debug-gym/pull/315
• Adding simple agent and option to call llm with tools by @MarcCote in https://github.com/microsoft/debug-gym/pull/312
• class level prompts look nicer actually, keep args prompt too by @sordonia in https://github.com/microsoft/debug-gym/pull/322
• Fix History Off-by-one Bug by @xingdi-eric-yuan in https://github.com/microsoft/debug-gym/pull/323
• Adding SWE-QA by @MarcCote in https://github.com/microsoft/debug-gym/pull/317
• Make sure workers write DEBUG log to file. by @matheper in https://github.com/microsoft/debug-gym/pull/324
• Timeouts by @xingdi-eric-yuan in https://github.com/microsoft/debug-gym/pull/325
• Security and Code Quality Improvements by @xingdi-eric-yuan in https://github.com/microsoft/debug-gym/pull/327
• Fix nohup commands not returning immediately with timeout wrapper by @xingdi-eric-yuan in https://github.com/microsoft/debug-gym/pull/329
• YAML by @xingdi-eric-yuan in https://github.com/microsoft/debug-gym/pull/328
• Bump werkzeug from 3.0.6 to 3.1.5 in /analysis/sft_data_viewer by @dependabot[bot] in https://github.com/microsoft/debug-gym/pull/331

New Contributors

• @dependabot[bot] made their first contribution in https://github.com/microsoft/debug-gym/pull/331

Full Changelog: https://github.com/microsoft/debug-gym/compare/1.2.0...1.3.0rc1