f-trycua · Oct 19, 2026 · Oct 19, 2026 · Oct 19, 2026 · Oct 19, 2026
@@ -198,8 +198,19 @@ def test_version_check_catches_a_package_file_that_drifted(tmp_path):
 
 
 def test_legacy_baselines_are_exactly_the_enrolled_packages_last_legacy_tags():
-    expected = {f"{component}-v{MANIFEST[path]}" for path, component in ENROLLED.items()}
-    assert release_notes.LEGACY_BASELINE_TAGS == expected
+    # One baseline per enrolled package, named for its last legacy tag. Until the
+    # package's first Release Please release that is the manifest version; the
+    # release PR then moves the manifest ahead of it (the baseline must stay for
+    # the notes of that release), so the baseline may be older, never newer.
+    by_component = {}
+    for tag in release_notes.LEGACY_BASELINE_TAGS:
+        component, version = tag.rsplit("-v", 1)
+        assert component not in by_component, f"two legacy baselines for {component}"
+        by_component[component] = tuple(int(part) for part in version.split("."))
+    assert set(by_component) == set(ENROLLED.values())
+    for path, component in ENROLLED.items():
+        released = tuple(int(part) for part in MANIFEST[path].split("."))
+        assert by_component[component] <= released, f"{component} baseline is newer than its manifest version"
 
 
 def test_only_a_legacy_baseline_relaxes_the_changelog_citation_check(monkeypatch, tmp_path):

@@ -8,7 +8,7 @@
   "apps/cua-spaces-macos": "0.10.0",
   "libs/python/agent": "0.9.0",
   "libs/python/core": "0.3.2",
-  "libs/cua-bench": "0.3.0",
+  "libs/cua-bench": "0.3.1",
   "libs/python/bench-ui": "0.7.1",
   "libs/python/som": "0.1.4",
   "libs/python/cua-sandbox-apps": "0.1.2",

diff --git a/docs/content/docs/cua-bench/reference/cli/agent.mdx b/docs/content/docs/cua-bench/reference/cli/agent.mdx
@@ -7,7 +7,7 @@ description: "Scaffold custom agents."
   AUTO-GENERATED FILE - DO NOT EDIT DIRECTLY
   Generated by: pnpm --dir docs docs:generate:cua-bench
   Source: scripts/docs-generators/extract_cb_docs.py (cua_bench)
-  Version: 0.3.0
+  Version: 0.3.1
 */}
 
 | Command | Description |

diff --git a/docs/content/docs/cua-bench/reference/cli/dataset.mdx b/docs/content/docs/cua-bench/reference/cli/dataset.mdx
@@ -7,7 +7,7 @@ description: "List registry datasets and build training data from runs."
   AUTO-GENERATED FILE - DO NOT EDIT DIRECTLY
   Generated by: pnpm --dir docs docs:generate:cua-bench
   Source: scripts/docs-generators/extract_cb_docs.py (cua_bench)
-  Version: 0.3.0
+  Version: 0.3.1
 */}
 
 | Command | Description |

diff --git a/docs/content/docs/cua-bench/reference/cli/env.mdx b/docs/content/docs/cua-bench/reference/cli/env.mdx
@@ -7,7 +7,7 @@ description: "Cloud pools and local sandboxes, sign-in, status and cleanup."
   AUTO-GENERATED FILE - DO NOT EDIT DIRECTLY
   Generated by: pnpm --dir docs docs:generate:cua-bench
   Source: scripts/docs-generators/extract_cb_docs.py (cua_bench)
-  Version: 0.3.0
+  Version: 0.3.1
 */}
 
 | Command | Description |

diff --git a/docs/content/docs/cua-bench/reference/cli/images.mdx b/docs/content/docs/cua-bench/reference/cli/images.mdx
@@ -7,7 +7,7 @@ description: "The canonical images and platform configurations."
   AUTO-GENERATED FILE - DO NOT EDIT DIRECTLY
   Generated by: pnpm --dir docs docs:generate:cua-bench
   Source: scripts/docs-generators/extract_cb_docs.py (cua_bench)
-  Version: 0.3.0
+  Version: 0.3.1
 */}
 
 | Command | Description |

diff --git a/docs/content/docs/cua-bench/reference/cli/index.mdx b/docs/content/docs/cua-bench/reference/cli/index.mdx
@@ -7,7 +7,7 @@ description: "cua-bench: run computer-use tasks and benchmark datasets on local
   AUTO-GENERATED FILE - DO NOT EDIT DIRECTLY
   Generated by: pnpm --dir docs docs:generate:cua-bench
   Source: scripts/docs-generators/extract_cb_docs.py (cua_bench)
-  Version: 0.3.0
+  Version: 0.3.1
 */}
 
 ```text output

diff --git a/docs/content/docs/cua-bench/reference/cli/run.mdx b/docs/content/docs/cua-bench/reference/cli/run.mdx
@@ -7,7 +7,7 @@ description: "Run a task or dataset with an agent or the oracle, run one interac
   AUTO-GENERATED FILE - DO NOT EDIT DIRECTLY
   Generated by: pnpm --dir docs docs:generate:cua-bench
   Source: scripts/docs-generators/extract_cb_docs.py (cua_bench)
-  Version: 0.3.0
+  Version: 0.3.1
 */}
 
 | Command | Description |

diff --git a/docs/content/docs/cua-bench/reference/cli/task.mdx b/docs/content/docs/cua-bench/reference/cli/task.mdx
@@ -7,7 +7,7 @@ description: "Inspect, list, scaffold and generate tasks."
   AUTO-GENERATED FILE - DO NOT EDIT DIRECTLY
   Generated by: pnpm --dir docs docs:generate:cua-bench
   Source: scripts/docs-generators/extract_cb_docs.py (cua_bench)
-  Version: 0.3.0
+  Version: 0.3.1
 */}
 
 | Command | Description |

diff --git a/docs/content/docs/cua-bench/reference/cli/trace.mdx b/docs/content/docs/cua-bench/reference/cli/trace.mdx
@@ -7,7 +7,7 @@ description: "View traces and agent trajectories."
   AUTO-GENERATED FILE - DO NOT EDIT DIRECTLY
   Generated by: pnpm --dir docs docs:generate:cua-bench
   Source: scripts/docs-generators/extract_cb_docs.py (cua_bench)
-  Version: 0.3.0
+  Version: 0.3.1
 */}
 
 | Command | Description |

diff --git a/docs/content/docs/cua-bench/reference/index.mdx b/docs/content/docs/cua-bench/reference/index.mdx
@@ -7,7 +7,7 @@ description: "Every cb command, the task module contract and the result files, g
   AUTO-GENERATED FILE - DO NOT EDIT DIRECTLY
   Generated by: pnpm --dir docs docs:generate:cua-bench
   Source: scripts/docs-generators/extract_cb_docs.py (cua_bench)
-  Version: 0.3.0
+  Version: 0.3.1
 */}
 
 ## Commands

diff --git a/docs/content/docs/cua-bench/reference/results.mdx b/docs/content/docs/cua-bench/reference/results.mdx
@@ -7,7 +7,7 @@ description: "The result.json and summary.json a run writes: every key, its type
   AUTO-GENERATED FILE - DO NOT EDIT DIRECTLY
   Generated by: pnpm --dir docs docs:generate:cua-bench
   Source: scripts/docs-generators/extract_cb_docs.py (cua_bench)
-  Version: 0.3.0
+  Version: 0.3.1
 */}
 
 A run writes one directory per variant and a `summary.json` next to them:

diff --git a/docs/content/docs/cua-bench/reference/task-definition.mdx b/docs/content/docs/cua-bench/reference/task-definition.mdx
@@ -7,7 +7,7 @@ description: "The task module contract: cb.Task, setup_config, the lifecycle dec
   AUTO-GENERATED FILE - DO NOT EDIT DIRECTLY
   Generated by: pnpm --dir docs docs:generate:cua-bench
   Source: scripts/docs-generators/extract_cb_docs.py (cua_bench)
-  Version: 0.3.0
+  Version: 0.3.1
 */}
 
 A task is a directory with a `main.py`. `cb` imports it, calls the `@cb.tasks_config` function to list the variants, then for each variant starts a sandbox from its `computer` and runs setup, the agent (or the `@cb.solve_task` oracle) and evaluation. Each lifecycle function receives the variant's `cb.Task` and the session on its sandbox.

diff --git a/libs/cua-bench-rl/uv.lock b/libs/cua-bench-rl/uv.lock
@@ -0,0 +1,14 @@
+# Changelog
+
+## [0.3.1](https://github.com/trycua/cua/compare/bench-v0.3.0...bench-v0.3.1) (2026-10-19)
+
+
+### Bug Fixes
+
+* **cua-bench:** fail fast on unsupported Windows Arena hosts ([#5356](https://github.com/trycua/cua/issues/5356)) ([2236b53](https://github.com/trycua/cua/commit/2236b5372a372067239a4b5f44efc595e0f8aca1))
+* **cua-bench:** pin bench-osworld to verified-20261003-fe6d89d ([#4580](https://github.com/trycua/cua/issues/4580)) ([94e7e9f](https://github.com/trycua/cua/commit/94e7e9fbc1b1bf467ff93f272dccc527f5d61dc6)), closes [#4519](https://github.com/trycua/cua/issues/4519)
+* **cua-bench:** raise when the sandbox does not answer in file_exists/directory_exists ([#5194](https://github.com/trycua/cua/issues/5194)) ([1ff5710](https://github.com/trycua/cua/commit/1ff57101e705d5964ca344e26207366a94467dee))
+* **cua-bench:** reject missing, nonfinite and nonnumeric evaluator rewards ([#5163](https://github.com/trycua/cua/issues/5163)) ([341d1e1](https://github.com/trycua/cua/commit/341d1e17237815340991376828992115a64ee15c))
+* lock the cua SDK checkout at 0.3.1 ([#4593](https://github.com/trycua/cua/issues/4593)) ([fcb386f](https://github.com/trycua/cua/commit/fcb386f4f958d1357af31ccb0056376f06d9db37))
+* **python:** cua-sandbox, cua-bench and the playground no longer depend on Fleet; remove cua-train (CUA-1834) ([#5445](https://github.com/trycua/cua/issues/5445)) ([7733289](https://github.com/trycua/cua/commit/773328941ead1b83a616e8b6c5a5ac4851af7114))
+* stop Cua Cloud (on="cloud") with a clear "Cua Cloud has ended" error (CUA-1854) ([#5415](https://github.com/trycua/cua/issues/5415)) ([5d78ec8](https://github.com/trycua/cua/commit/5d78ec8b7705d942a3d260b276bcce76614a634c))
@@ -1,6 +1,6 @@
 [project]
 name = "cua-bench"
-version = "0.3.0"
+version = "0.3.1"
 description = "Run computer-use benchmark datasets on local and cloud sandboxes"
 readme = "README.md"
 license = "MIT"

diff --git a/scripts/docs-generators/cli-specs/cb.json b/scripts/docs-generators/cli-specs/cb.json
@@ -2771,5 +2771,5 @@
  "global_options": [],
  "name": "cb",
  "usage": "cb <COMMAND> [OPTIONS]",
- "version": "0.3.0"
+ "version": "0.3.1"
 }