breken
Try breken for free
7 days free
Breken maps your codebase, proves what is dead and what is broken, and opens the pull request that fixes it.
Sign up with GitHubNo card required. You pick the repositories. We read and write only those you grant.
Self-Healing Software
manifestoBacked by
- Good catch, fix is correct and well tested.ericcurtin · docker/model-runner#1068
- @breken thank you for the analysis and the fix. I investigated #15006 and this PR against release-1.12.2, and I can confirm both the bug and the fix. …erichare · langflow-ai/langflow#15007
- Thank you for this, @breken! Your diagnosis was exactly right: disk_ann only exists in the opensearch-jvector plugin, so choosing the method from the engine (disk_ann for jvector, hnsw for nmslib / faiss / lucene) is the correct fix. We confirmed it against stock OpenSearch 2.19.3 and 3.8.0. …erichare · langflow-ai/langflow#15095
- @breken heads-up: this PR's branch has been rebased onto release-1.12.2 and force-pushed to fix-15006-check-constraint-naming on your fork, and the PR base is now release-1.12.2. …erichare · langflow-ai/langflow#15007
- Manually merged with amends to #1207 (cc @arjunpkulkarni)v0idpwn · supabase/supavisor#1209
- Confirmed that cell_text in studio/backend/utils/datasets/cells.py still falls back to the dict repr for a SQuAD answers cell, while AI Assist already unwraps it. Will get this reviewed.danielhanchen · unslothai/unsloth#12056
- Confirmed that visible_text in unsloth_cli/_inference.py still holds back a trailing < or <th after the stream ends, so it is dropped from the output and the chat history. Will get this reviewed.danielhanchen · unslothai/unsloth#11893
- Confirmed main still tests cells for truthiness in dataset_utils.py and chat_templates.py, so a 0 label trains an empty reply and NaN trains "nan" on the chat mapping paths. Since this touches chat_templates.py, could you confirm the rendered text is unchanged for rows whose cells are ordinary non-empty strings? …danielhanchen · unslothai/unsloth#11895
- The fix looks nice, but it misses tests and documentation. I have addressed this on this PR: https://github.com/penpot/penpot/pull/11906niwinz · penpot/penpot#11520
- I merging this as is and the, rebase the other. Thanksniwinz · penpot/penpot#11520
- Thanks for your contribution! We've received your pull request and assigned it for review. We have a lot on our plate right now, so it may take us a little while to get to it but we appreciate your patience and will be in touch as soon as we can!madalenapmelo-kp · penpot/penpot#11520
- Some tests fail - rest is LGTMbircni · go-gitea/gitea#39354
- Since it's an external repository, we can't get that at the moment.lunny · go-gitea/gitea#39343
- Do you have it?wxiaoguang · go-gitea/gitea#39343
- BTW I think we should probably replace all "page flash after request" with real toasts which don't need such intrusive scrolling.silverwind · go-gitea/gitea#39354
- That's impossible at the moment. Many pages depend on "full reload" (flash message + redirect) behavior.wxiaoguang · go-gitea/gitea#39354
- Right, full page reloads should go as well where possible. Could show a simple "Success" toast after saving and update the page live. But maybe this needs a framework first.silverwind · go-gitea/gitea#39354
- Yep, we can keep improving the "message" framework in the future. The current design is flexible enough and won't block future improvements.wxiaoguang · go-gitea/gitea#39354
- Thank you for the contribution @breken !sony-mathew · chatwoot/chatwoot#15820
- @breken Thanks for the contribution. I removed the initializer and kept the environment-variable documentation: sidekiq_alive 2.5.0 already reads SIDEKIQ_ALIVE_HOST and SIDEKIQ_ALIVE_PORT directly in its configuration defaults, and its server configuration converts the port to an integer. …sony-mathew · chatwoot/chatwoot#15820
- lgtmsony-mathew · chatwoot/chatwoot#15820
- @breken Thank you for your contribution to Twenty. We appreciate it.neo773 · twentyhq/twenty#25751
- Thank you!charlesBochet · twentyhq/twenty#25828
- Shipped in v0.9.62 (now on PyPI: uv tool install graphifyy==0.9.62). Your commits were cherry-picked with authorship preserved, so this shows up under your GitHub contributions. Thanks @breken!safishamsi · Graphify-Labs/graphify#3546
- Shipped in v0.9.68 (now on PyPI) via an authorship-preserving cherry-pick, so your commit keeps contributor-graph credit. Thanks @breken! Terraform secret-named variable-default/output redaction. Release: https://github.com/Graphify-Labs/graphify/releases/tag/v0.9.68safishamsi · Graphify-Labs/graphify#3817
- Good catch. "last update:" ending in "date:" really did swallow the fact, so the engine went in with no date at all. Merging.HenryNdubuaku · cactus-compute/needle#151
- Merges clean and the suite passes with it, including your new cases. Thanks.HenryNdubuaku · cactus-compute/needle#126
- Yes, extract() should check the call against the same facts the engine was given, not the caller's raw system string. Merging.HenryNdubuaku · cactus-compute/needle#152
- Reproduced on main: _source_years("Book a table on 5 June 19:30") gives {19}, so no ISO date could ever match. Your fix plus the four patterns in the test are right. Merging. Your note about the 3.0.1 engine flagging every date argument is a separate and bigger problem, and I am looking at it now. …HenryNdubuaku · cactus-compute/needle#154
- Following up on your note that the 3.0.1 engine flagged every date argument by itself. …HenryNdubuaku · cactus-compute/needle#154
- thanks for your contribution!injaneity · trycua/cua#4134
- Thanks for your contribution! @breken, please sign the Contributor License Agreement (CLA) and resolve the conflicts in this PR.mrpositron · lightly-ai/lightly-studio#2310
- Cool, thanks!collindutter · griptape-ai/griptape#2335
- the unit test is sufficient, ditch thisshikhar · s2-streamstore/s2#781
- @breken do u iterateshikhar · s2-streamstore/s2#781
- Reviewed — looks correct to me (not a formal approval; storage isn't my ownership area, so leaving the merge decision to @doobidoo). Verified the three fixes against the sqlite-vec reference implementation: 1. …filhocf · doobidoo/mcp-memory-service#1214
- Thanks @breken, merged! The regression tests catch the wrong score as well as the changed reducer, which makes the impact of this bug clear. Nice focused fix.jeqcho · robocurve/inspect-robots#447
- @breken great catch and thanks for the PR! mergingguyb1 · onecli/onecli#559
- LGTM, solid regression test. Follow-ups: submodules also use a .git file pointing at a gitdir under .git/modules/, worth the same coverage; and the full container-plan assertion is skipped on Windows only because mount_path rejects drive letters, worth revisiting once that's lifted.ericcurtin · llmmanorg/llmman#560
- Security gap: git_metadata_paths only re-validates when .git is a gitdir: pointer file. If .git is instead a directory symlink (e.g. ln -s /.ssh gets mounted read-only straight into the guest at workspace/.git. Verified locally: plan() succeeds and produces that mount. …ericcurtin · llmmanorg/llmman#561
- Marking as draft: the symlinked-.git bypass above still leaks host directories read-only into the sandbox. Please reopen for review once that's closed.ericcurtin · llmmanorg/llmman#561
- LGTMricky-chaoju · llmmanorg/llmman#561
- Nice, thorough fix. One small thought for later: the index-or-position fallback now exists three times (here, responses.rs, messages.rs). Might be worth a shared helper someday, not blocking this PR.ericcurtin · llmmanorg/llmman#557
- Nit: the let git = |cwd, args| {...} closure is now duplicated 4x across tests in this file. Worth extracting to a shared test helper (e.g. fn run_git).ericcurtin · llmmanorg/llmman#561
- Good catch, fix is correct and well tested.ericcurtin · docker/model-runner#1068
- @breken thank you for the analysis and the fix. I investigated #15006 and this PR against release-1.12.2, and I can confirm both the bug and the fix. …erichare · langflow-ai/langflow#15007
- Thank you for this, @breken! Your diagnosis was exactly right: disk_ann only exists in the opensearch-jvector plugin, so choosing the method from the engine (disk_ann for jvector, hnsw for nmslib / faiss / lucene) is the correct fix. We confirmed it against stock OpenSearch 2.19.3 and 3.8.0. …erichare · langflow-ai/langflow#15095
- @breken heads-up: this PR's branch has been rebased onto release-1.12.2 and force-pushed to fix-15006-check-constraint-naming on your fork, and the PR base is now release-1.12.2. …erichare · langflow-ai/langflow#15007
- Manually merged with amends to #1207 (cc @arjunpkulkarni)v0idpwn · supabase/supavisor#1209
- Confirmed that cell_text in studio/backend/utils/datasets/cells.py still falls back to the dict repr for a SQuAD answers cell, while AI Assist already unwraps it. Will get this reviewed.danielhanchen · unslothai/unsloth#12056
- Confirmed that visible_text in unsloth_cli/_inference.py still holds back a trailing < or <th after the stream ends, so it is dropped from the output and the chat history. Will get this reviewed.danielhanchen · unslothai/unsloth#11893
- Confirmed main still tests cells for truthiness in dataset_utils.py and chat_templates.py, so a 0 label trains an empty reply and NaN trains "nan" on the chat mapping paths. Since this touches chat_templates.py, could you confirm the rendered text is unchanged for rows whose cells are ordinary non-empty strings? …danielhanchen · unslothai/unsloth#11895
- The fix looks nice, but it misses tests and documentation. I have addressed this on this PR: https://github.com/penpot/penpot/pull/11906niwinz · penpot/penpot#11520
- I merging this as is and the, rebase the other. Thanksniwinz · penpot/penpot#11520
- Thanks for your contribution! We've received your pull request and assigned it for review. We have a lot on our plate right now, so it may take us a little while to get to it but we appreciate your patience and will be in touch as soon as we can!madalenapmelo-kp · penpot/penpot#11520
- Some tests fail - rest is LGTMbircni · go-gitea/gitea#39354
- Since it's an external repository, we can't get that at the moment.lunny · go-gitea/gitea#39343
- Do you have it?wxiaoguang · go-gitea/gitea#39343
- BTW I think we should probably replace all "page flash after request" with real toasts which don't need such intrusive scrolling.silverwind · go-gitea/gitea#39354
- That's impossible at the moment. Many pages depend on "full reload" (flash message + redirect) behavior.wxiaoguang · go-gitea/gitea#39354
- Right, full page reloads should go as well where possible. Could show a simple "Success" toast after saving and update the page live. But maybe this needs a framework first.silverwind · go-gitea/gitea#39354
- Yep, we can keep improving the "message" framework in the future. The current design is flexible enough and won't block future improvements.wxiaoguang · go-gitea/gitea#39354
- Thank you for the contribution @breken !sony-mathew · chatwoot/chatwoot#15820
- @breken Thanks for the contribution. I removed the initializer and kept the environment-variable documentation: sidekiq_alive 2.5.0 already reads SIDEKIQ_ALIVE_HOST and SIDEKIQ_ALIVE_PORT directly in its configuration defaults, and its server configuration converts the port to an integer. …sony-mathew · chatwoot/chatwoot#15820
- lgtmsony-mathew · chatwoot/chatwoot#15820
- @breken Thank you for your contribution to Twenty. We appreciate it.neo773 · twentyhq/twenty#25751
- Thank you!charlesBochet · twentyhq/twenty#25828
- Shipped in v0.9.62 (now on PyPI: uv tool install graphifyy==0.9.62). Your commits were cherry-picked with authorship preserved, so this shows up under your GitHub contributions. Thanks @breken!safishamsi · Graphify-Labs/graphify#3546
- Shipped in v0.9.68 (now on PyPI) via an authorship-preserving cherry-pick, so your commit keeps contributor-graph credit. Thanks @breken! Terraform secret-named variable-default/output redaction. Release: https://github.com/Graphify-Labs/graphify/releases/tag/v0.9.68safishamsi · Graphify-Labs/graphify#3817
- Good catch. "last update:" ending in "date:" really did swallow the fact, so the engine went in with no date at all. Merging.HenryNdubuaku · cactus-compute/needle#151
- Merges clean and the suite passes with it, including your new cases. Thanks.HenryNdubuaku · cactus-compute/needle#126
- Yes, extract() should check the call against the same facts the engine was given, not the caller's raw system string. Merging.HenryNdubuaku · cactus-compute/needle#152
- Reproduced on main: _source_years("Book a table on 5 June 19:30") gives {19}, so no ISO date could ever match. Your fix plus the four patterns in the test are right. Merging. Your note about the 3.0.1 engine flagging every date argument is a separate and bigger problem, and I am looking at it now. …HenryNdubuaku · cactus-compute/needle#154
- Following up on your note that the 3.0.1 engine flagged every date argument by itself. …HenryNdubuaku · cactus-compute/needle#154
- thanks for your contribution!injaneity · trycua/cua#4134
- Thanks for your contribution! @breken, please sign the Contributor License Agreement (CLA) and resolve the conflicts in this PR.mrpositron · lightly-ai/lightly-studio#2310
- Cool, thanks!collindutter · griptape-ai/griptape#2335
- the unit test is sufficient, ditch thisshikhar · s2-streamstore/s2#781
- @breken do u iterateshikhar · s2-streamstore/s2#781
- Reviewed — looks correct to me (not a formal approval; storage isn't my ownership area, so leaving the merge decision to @doobidoo). Verified the three fixes against the sqlite-vec reference implementation: 1. …filhocf · doobidoo/mcp-memory-service#1214
- Thanks @breken, merged! The regression tests catch the wrong score as well as the changed reducer, which makes the impact of this bug clear. Nice focused fix.jeqcho · robocurve/inspect-robots#447
- @breken great catch and thanks for the PR! mergingguyb1 · onecli/onecli#559
- LGTM, solid regression test. Follow-ups: submodules also use a .git file pointing at a gitdir under .git/modules/, worth the same coverage; and the full container-plan assertion is skipped on Windows only because mount_path rejects drive letters, worth revisiting once that's lifted.ericcurtin · llmmanorg/llmman#560
- Security gap: git_metadata_paths only re-validates when .git is a gitdir: pointer file. If .git is instead a directory symlink (e.g. ln -s /.ssh gets mounted read-only straight into the guest at workspace/.git. Verified locally: plan() succeeds and produces that mount. …ericcurtin · llmmanorg/llmman#561
- Marking as draft: the symlinked-.git bypass above still leaks host directories read-only into the sandbox. Please reopen for review once that's closed.ericcurtin · llmmanorg/llmman#561
- LGTMricky-chaoju · llmmanorg/llmman#561
- Nice, thorough fix. One small thought for later: the index-or-position fallback now exists three times (here, responses.rs, messages.rs). Might be worth a shared helper someday, not blocking this PR.ericcurtin · llmmanorg/llmman#557
- Nit: the let git = |cwd, args| {...} closure is now duplicated 4x across tests in this file. Worth extracting to a shared test helper (e.g. fn run_git).ericcurtin · llmmanorg/llmman#561