fix(test): repair four unit tests drifting behind source and upstream APIs

[jariy17]

Draft. Depends on #609 — see Ordering below.

Fixes the 4 pre-existing unit-test failures on main that #609 does not cover. All four are tests (or a message/doc) that fell behind source and upstream changes; none are product bugs.

1 + 2. TestAsyncMode multi-agent and bidi callbacks

test_agentcore_memory_session_manager.py:3726 and :3756 reached into the private registry dict:

callbacks=registry._registered_callbacks.get(event_type, [])
assertall(asyncio.iscoroutinefunction(cb) forcbincallbacks) # always False

Upstream Strands now stores callbacks wrapped in a _CallbackEntry:

type= <class 'strands.hooks.registry._CallbackEntry'>
repr= _CallbackEntry(callback=<function ...register_hooks.<locals>._offload.<locals>._callback ...>)

iscoroutinefunction() on the wrapper is always False, so the assertion could never hold. The source is correct — session_manager.py:989-991 does register async _offload callbacks.

The fix is the accessor the passing sibling tests already use (lines 3636–3714): public registry.get_callbacks_for(event), which unwraps.

Side benefit: the bidi-init assertion at :3751 is not any(iscoroutinefunction(...)). It was passing for the wrong reason — the wrapper is never a coroutine function, so it would have passed even if the callback were wrongly async. Against unwrapped callbacks it now actually tests what it claims.

3. EvaluatorOutput validator message — source fix

The message advertised a value escape hatch that does not exist:

ifnotself.errorCodeandself.labelisNone: # value is never inspectedraiseValueError("Either label, value, or errorCode must be set; ...")

The class docstring (models.py:63-65) and every caller in src/ agree that label is required unless errorCode is set, so the message was the defect, not the logic. Corrected to "label is required for success responses; ...", which is also what the test expected.

4. test_evaluation_with_empty_trajectory

TypeError: 'EvaluationReport' object is not subscriptable

run_evaluations() returns a single report now, verified against the installed library:

run_evaluations -> <class 'strands_evals.types.evaluation_report.EvaluationReport'>
model_fields: ['overall_score', 'scores', 'cases', 'test_passes', ...]

Dropped the [0]. Also fixed the same stale pattern in README.md (2 occurrences) — the documented reports = ...; report = reports[0] snippet would hand users the identical TypeError.

Testing

• uv run pytest tests/ on Python 3.10 → 2934 passed, 10 skipped, 4 xpassed, 0 failed
• TestAsyncMode in isolation → 8 passed
• ruff check src/ tests/ → all checks passed; ruff format applied (the shortened message now fits one line — verified main was format-clean beforehand, so that reflow is mine)

Ordering

This branch is cut from main, so on its own CI still hits the 2 deepeval collection errors that #609 fixes — verified with a clean uv sync --dev:

ERROR tests/.../third_party/deepeval/test_adapter.py
ERROR tests/.../third_party/deepeval/test_error_handling.py
Interrupted: 2 errors during collection

Kept as a draft for that reason. #609 then this turns ci.yml green; merge order the other way leaves 4 failures. Happy to rebase on #609 if you'd prefer to see this one green before review.

Related: #608 (integration workflow), #609 (dev-group pin).

[github-actions]

✅ No Breaking Changes Detected

No public API breaking changes found in this PR.

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)

[jariy17]

Draft. Depends on #609 — see Ordering below.

Fixes the 4 pre-existing unit-test failures on main that #609 does not cover. All four are tests (or a message/doc) that fell behind source and upstream changes; none are product bugs.

1 + 2. TestAsyncMode multi-agent and bidi callbacks

test_agentcore_memory_session_manager.py:3726 and :3756 reached into the private registry dict:

callbacks=registry._registered_callbacks.get(event_type, [])
assertall(asyncio.iscoroutinefunction(cb) forcbincallbacks) # always False

Upstream Strands now stores callbacks wrapped in a _CallbackEntry:

type= <class 'strands.hooks.registry._CallbackEntry'>
repr= _CallbackEntry(callback=<function ...register_hooks.<locals>._offload.<locals>._callback ...>)

iscoroutinefunction() on the wrapper is always False, so the assertion could never hold. The source is correct — session_manager.py:989-991 does register async _offload callbacks.

The fix is the accessor the passing sibling tests already use (lines 3636–3714): public registry.get_callbacks_for(event), which unwraps.

Side benefit: the bidi-init assertion at :3751 is not any(iscoroutinefunction(...)). It was passing for the wrong reason — the wrapper is never a coroutine function, so it would have passed even if the callback were wrongly async. Against unwrapped callbacks it now actually tests what it claims.

3. EvaluatorOutput validator message — source fix

The message advertised a value escape hatch that does not exist:

ifnotself.errorCodeandself.labelisNone: # value is never inspectedraiseValueError("Either label, value, or errorCode must be set; ...")

The class docstring (models.py:63-65) and every caller in src/ agree that label is required unless errorCode is set, so the message was the defect, not the logic. Corrected to "label is required for success responses; ...", which is also what the test expected.

4. test_evaluation_with_empty_trajectory

TypeError: 'EvaluationReport' object is not subscriptable

run_evaluations() returns a single report now, verified against the installed library:

run_evaluations -> <class 'strands_evals.types.evaluation_report.EvaluationReport'>
model_fields: ['overall_score', 'scores', 'cases', 'test_passes', ...]

Dropped the [0]. Also fixed the same stale pattern in README.md (2 occurrences) — the documented reports = ...; report = reports[0] snippet would hand users the identical TypeError.

Testing

• uv run pytest tests/ on Python 3.10 → 2934 passed, 10 skipped, 4 xpassed, 0 failed
• TestAsyncMode in isolation → 8 passed
• ruff check src/ tests/ → all checks passed; ruff format applied (the shortened message now fits one line — verified main was format-clean beforehand, so that reflow is mine)

Ordering

This branch is cut from main, so on its own CI still hits the 2 deepeval collection errors that #609 fixes — verified with a clean uv sync --dev:

ERROR tests/.../third_party/deepeval/test_adapter.py
ERROR tests/.../third_party/deepeval/test_error_handling.py
Interrupted: 2 errors during collection

Kept as a draft for that reason. #609 then this turns ci.yml green; merge order the other way leaves 4 failures. Happy to rebase on #609 if you'd prefer to see this one green before review.

Related: #608 (integration workflow), #609 (dev-group pin).

[github-actions]

✅ No Breaking Changes Detected

No public API breaking changes found in this PR.

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)

[jariy17]

Draft. Depends on #609 — see Ordering below.

Fixes the 4 pre-existing unit-test failures on main that #609 does not cover. All four are tests (or a message/doc) that fell behind source and upstream changes; none are product bugs.

1 + 2. TestAsyncMode multi-agent and bidi callbacks

test_agentcore_memory_session_manager.py:3726 and :3756 reached into the private registry dict:

callbacks=registry._registered_callbacks.get(event_type, [])
assertall(asyncio.iscoroutinefunction(cb) forcbincallbacks) # always False

Upstream Strands now stores callbacks wrapped in a _CallbackEntry:

type= <class 'strands.hooks.registry._CallbackEntry'>
repr= _CallbackEntry(callback=<function ...register_hooks.<locals>._offload.<locals>._callback ...>)

iscoroutinefunction() on the wrapper is always False, so the assertion could never hold. The source is correct — session_manager.py:989-991 does register async _offload callbacks.

The fix is the accessor the passing sibling tests already use (lines 3636–3714): public registry.get_callbacks_for(event), which unwraps.

Side benefit: the bidi-init assertion at :3751 is not any(iscoroutinefunction(...)). It was passing for the wrong reason — the wrapper is never a coroutine function, so it would have passed even if the callback were wrongly async. Against unwrapped callbacks it now actually tests what it claims.

3. EvaluatorOutput validator message — source fix

The message advertised a value escape hatch that does not exist:

ifnotself.errorCodeandself.labelisNone: # value is never inspectedraiseValueError("Either label, value, or errorCode must be set; ...")

The class docstring (models.py:63-65) and every caller in src/ agree that label is required unless errorCode is set, so the message was the defect, not the logic. Corrected to "label is required for success responses; ...", which is also what the test expected.

4. test_evaluation_with_empty_trajectory

TypeError: 'EvaluationReport' object is not subscriptable

run_evaluations() returns a single report now, verified against the installed library:

run_evaluations -> <class 'strands_evals.types.evaluation_report.EvaluationReport'>
model_fields: ['overall_score', 'scores', 'cases', 'test_passes', ...]

Dropped the [0]. Also fixed the same stale pattern in README.md (2 occurrences) — the documented reports = ...; report = reports[0] snippet would hand users the identical TypeError.

Testing

• uv run pytest tests/ on Python 3.10 → 2934 passed, 10 skipped, 4 xpassed, 0 failed
• TestAsyncMode in isolation → 8 passed
• ruff check src/ tests/ → all checks passed; ruff format applied (the shortened message now fits one line — verified main was format-clean beforehand, so that reflow is mine)

Ordering

This branch is cut from main, so on its own CI still hits the 2 deepeval collection errors that #609 fixes — verified with a clean uv sync --dev:

ERROR tests/.../third_party/deepeval/test_adapter.py
ERROR tests/.../third_party/deepeval/test_error_handling.py
Interrupted: 2 errors during collection

Kept as a draft for that reason. #609 then this turns ci.yml green; merge order the other way leaves 4 failures. Happy to rebase on #609 if you'd prefer to see this one green before review.

Related: #608 (integration workflow), #609 (dev-group pin).

[github-actions]

✅ No Breaking Changes Detected

No public API breaking changes found in this PR.

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)

[jariy17]

Draft. Depends on #609 — see Ordering below.

Fixes the 4 pre-existing unit-test failures on main that #609 does not cover. All four are tests (or a message/doc) that fell behind source and upstream changes; none are product bugs.

1 + 2. TestAsyncMode multi-agent and bidi callbacks

test_agentcore_memory_session_manager.py:3726 and :3756 reached into the private registry dict:

callbacks=registry._registered_callbacks.get(event_type, [])
assertall(asyncio.iscoroutinefunction(cb) forcbincallbacks) # always False

Upstream Strands now stores callbacks wrapped in a _CallbackEntry:

type= <class 'strands.hooks.registry._CallbackEntry'>
repr= _CallbackEntry(callback=<function ...register_hooks.<locals>._offload.<locals>._callback ...>)

iscoroutinefunction() on the wrapper is always False, so the assertion could never hold. The source is correct — session_manager.py:989-991 does register async _offload callbacks.

The fix is the accessor the passing sibling tests already use (lines 3636–3714): public registry.get_callbacks_for(event), which unwraps.

Side benefit: the bidi-init assertion at :3751 is not any(iscoroutinefunction(...)). It was passing for the wrong reason — the wrapper is never a coroutine function, so it would have passed even if the callback were wrongly async. Against unwrapped callbacks it now actually tests what it claims.

3. EvaluatorOutput validator message — source fix

The message advertised a value escape hatch that does not exist:

ifnotself.errorCodeandself.labelisNone: # value is never inspectedraiseValueError("Either label, value, or errorCode must be set; ...")

The class docstring (models.py:63-65) and every caller in src/ agree that label is required unless errorCode is set, so the message was the defect, not the logic. Corrected to "label is required for success responses; ...", which is also what the test expected.

4. test_evaluation_with_empty_trajectory

TypeError: 'EvaluationReport' object is not subscriptable

run_evaluations() returns a single report now, verified against the installed library:

run_evaluations -> <class 'strands_evals.types.evaluation_report.EvaluationReport'>
model_fields: ['overall_score', 'scores', 'cases', 'test_passes', ...]

Dropped the [0]. Also fixed the same stale pattern in README.md (2 occurrences) — the documented reports = ...; report = reports[0] snippet would hand users the identical TypeError.

Testing

• uv run pytest tests/ on Python 3.10 → 2934 passed, 10 skipped, 4 xpassed, 0 failed
• TestAsyncMode in isolation → 8 passed
• ruff check src/ tests/ → all checks passed; ruff format applied (the shortened message now fits one line — verified main was format-clean beforehand, so that reflow is mine)

Ordering

This branch is cut from main, so on its own CI still hits the 2 deepeval collection errors that #609 fixes — verified with a clean uv sync --dev:

ERROR tests/.../third_party/deepeval/test_adapter.py
ERROR tests/.../third_party/deepeval/test_error_handling.py
Interrupted: 2 errors during collection

Kept as a draft for that reason. #609 then this turns ci.yml green; merge order the other way leaves 4 failures. Happy to rebase on #609 if you'd prefer to see this one green before review.

Related: #608 (integration workflow), #609 (dev-group pin).

[github-actions]

✅ No Breaking Changes Detected

No public API breaking changes found in this PR.

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)

[jariy17]

Draft. Depends on #609 — see Ordering below.

Fixes the 4 pre-existing unit-test failures on main that #609 does not cover. All four are tests (or a message/doc) that fell behind source and upstream changes; none are product bugs.

1 + 2. TestAsyncMode multi-agent and bidi callbacks

test_agentcore_memory_session_manager.py:3726 and :3756 reached into the private registry dict:

callbacks=registry._registered_callbacks.get(event_type, [])
assertall(asyncio.iscoroutinefunction(cb) forcbincallbacks) # always False

Upstream Strands now stores callbacks wrapped in a _CallbackEntry:

type= <class 'strands.hooks.registry._CallbackEntry'>
repr= _CallbackEntry(callback=<function ...register_hooks.<locals>._offload.<locals>._callback ...>)

iscoroutinefunction() on the wrapper is always False, so the assertion could never hold. The source is correct — session_manager.py:989-991 does register async _offload callbacks.

The fix is the accessor the passing sibling tests already use (lines 3636–3714): public registry.get_callbacks_for(event), which unwraps.

Side benefit: the bidi-init assertion at :3751 is not any(iscoroutinefunction(...)). It was passing for the wrong reason — the wrapper is never a coroutine function, so it would have passed even if the callback were wrongly async. Against unwrapped callbacks it now actually tests what it claims.

3. EvaluatorOutput validator message — source fix

The message advertised a value escape hatch that does not exist:

ifnotself.errorCodeandself.labelisNone: # value is never inspectedraiseValueError("Either label, value, or errorCode must be set; ...")

The class docstring (models.py:63-65) and every caller in src/ agree that label is required unless errorCode is set, so the message was the defect, not the logic. Corrected to "label is required for success responses; ...", which is also what the test expected.

4. test_evaluation_with_empty_trajectory

TypeError: 'EvaluationReport' object is not subscriptable

run_evaluations() returns a single report now, verified against the installed library:

run_evaluations -> <class 'strands_evals.types.evaluation_report.EvaluationReport'>
model_fields: ['overall_score', 'scores', 'cases', 'test_passes', ...]

Dropped the [0]. Also fixed the same stale pattern in README.md (2 occurrences) — the documented reports = ...; report = reports[0] snippet would hand users the identical TypeError.

Testing

• uv run pytest tests/ on Python 3.10 → 2934 passed, 10 skipped, 4 xpassed, 0 failed
• TestAsyncMode in isolation → 8 passed
• ruff check src/ tests/ → all checks passed; ruff format applied (the shortened message now fits one line — verified main was format-clean beforehand, so that reflow is mine)

Ordering

This branch is cut from main, so on its own CI still hits the 2 deepeval collection errors that #609 fixes — verified with a clean uv sync --dev:

ERROR tests/.../third_party/deepeval/test_adapter.py
ERROR tests/.../third_party/deepeval/test_error_handling.py
Interrupted: 2 errors during collection

Kept as a draft for that reason. #609 then this turns ci.yml green; merge order the other way leaves 4 failures. Happy to rebase on #609 if you'd prefer to see this one green before review.

Related: #608 (integration workflow), #609 (dev-group pin).

[github-actions]

✅ No Breaking Changes Detected

No public API breaking changes found in this PR.

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)

[jariy17]

Draft. Depends on #609 — see Ordering below.

Fixes the 4 pre-existing unit-test failures on main that #609 does not cover. All four are tests (or a message/doc) that fell behind source and upstream changes; none are product bugs.

1 + 2. TestAsyncMode multi-agent and bidi callbacks

test_agentcore_memory_session_manager.py:3726 and :3756 reached into the private registry dict:

callbacks=registry._registered_callbacks.get(event_type, [])
assertall(asyncio.iscoroutinefunction(cb) forcbincallbacks) # always False

Upstream Strands now stores callbacks wrapped in a _CallbackEntry:

type= <class 'strands.hooks.registry._CallbackEntry'>
repr= _CallbackEntry(callback=<function ...register_hooks.<locals>._offload.<locals>._callback ...>)

iscoroutinefunction() on the wrapper is always False, so the assertion could never hold. The source is correct — session_manager.py:989-991 does register async _offload callbacks.

The fix is the accessor the passing sibling tests already use (lines 3636–3714): public registry.get_callbacks_for(event), which unwraps.

Side benefit: the bidi-init assertion at :3751 is not any(iscoroutinefunction(...)). It was passing for the wrong reason — the wrapper is never a coroutine function, so it would have passed even if the callback were wrongly async. Against unwrapped callbacks it now actually tests what it claims.

3. EvaluatorOutput validator message — source fix

The message advertised a value escape hatch that does not exist:

ifnotself.errorCodeandself.labelisNone: # value is never inspectedraiseValueError("Either label, value, or errorCode must be set; ...")

The class docstring (models.py:63-65) and every caller in src/ agree that label is required unless errorCode is set, so the message was the defect, not the logic. Corrected to "label is required for success responses; ...", which is also what the test expected.

4. test_evaluation_with_empty_trajectory

TypeError: 'EvaluationReport' object is not subscriptable

run_evaluations() returns a single report now, verified against the installed library:

run_evaluations -> <class 'strands_evals.types.evaluation_report.EvaluationReport'>
model_fields: ['overall_score', 'scores', 'cases', 'test_passes', ...]

Dropped the [0]. Also fixed the same stale pattern in README.md (2 occurrences) — the documented reports = ...; report = reports[0] snippet would hand users the identical TypeError.

Testing

• uv run pytest tests/ on Python 3.10 → 2934 passed, 10 skipped, 4 xpassed, 0 failed
• TestAsyncMode in isolation → 8 passed
• ruff check src/ tests/ → all checks passed; ruff format applied (the shortened message now fits one line — verified main was format-clean beforehand, so that reflow is mine)

Ordering

This branch is cut from main, so on its own CI still hits the 2 deepeval collection errors that #609 fixes — verified with a clean uv sync --dev:

ERROR tests/.../third_party/deepeval/test_adapter.py
ERROR tests/.../third_party/deepeval/test_error_handling.py
Interrupted: 2 errors during collection

Kept as a draft for that reason. #609 then this turns ci.yml green; merge order the other way leaves 4 failures. Happy to rebase on #609 if you'd prefer to see this one green before review.

Related: #608 (integration workflow), #609 (dev-group pin).

[github-actions]

✅ No Breaking Changes Detected

No public API breaking changes found in this PR.

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)

[jariy17]

Draft. Depends on #609 — see Ordering below.

Fixes the 4 pre-existing unit-test failures on main that #609 does not cover. All four are tests (or a message/doc) that fell behind source and upstream changes; none are product bugs.

1 + 2. TestAsyncMode multi-agent and bidi callbacks

test_agentcore_memory_session_manager.py:3726 and :3756 reached into the private registry dict:

callbacks=registry._registered_callbacks.get(event_type, [])
assertall(asyncio.iscoroutinefunction(cb) forcbincallbacks) # always False

Upstream Strands now stores callbacks wrapped in a _CallbackEntry:

type= <class 'strands.hooks.registry._CallbackEntry'>
repr= _CallbackEntry(callback=<function ...register_hooks.<locals>._offload.<locals>._callback ...>)

iscoroutinefunction() on the wrapper is always False, so the assertion could never hold. The source is correct — session_manager.py:989-991 does register async _offload callbacks.

The fix is the accessor the passing sibling tests already use (lines 3636–3714): public registry.get_callbacks_for(event), which unwraps.

Side benefit: the bidi-init assertion at :3751 is not any(iscoroutinefunction(...)). It was passing for the wrong reason — the wrapper is never a coroutine function, so it would have passed even if the callback were wrongly async. Against unwrapped callbacks it now actually tests what it claims.

3. EvaluatorOutput validator message — source fix

The message advertised a value escape hatch that does not exist:

ifnotself.errorCodeandself.labelisNone: # value is never inspectedraiseValueError("Either label, value, or errorCode must be set; ...")

The class docstring (models.py:63-65) and every caller in src/ agree that label is required unless errorCode is set, so the message was the defect, not the logic. Corrected to "label is required for success responses; ...", which is also what the test expected.

4. test_evaluation_with_empty_trajectory

TypeError: 'EvaluationReport' object is not subscriptable

run_evaluations() returns a single report now, verified against the installed library:

run_evaluations -> <class 'strands_evals.types.evaluation_report.EvaluationReport'>
model_fields: ['overall_score', 'scores', 'cases', 'test_passes', ...]

Dropped the [0]. Also fixed the same stale pattern in README.md (2 occurrences) — the documented reports = ...; report = reports[0] snippet would hand users the identical TypeError.

Testing

• uv run pytest tests/ on Python 3.10 → 2934 passed, 10 skipped, 4 xpassed, 0 failed
• TestAsyncMode in isolation → 8 passed
• ruff check src/ tests/ → all checks passed; ruff format applied (the shortened message now fits one line — verified main was format-clean beforehand, so that reflow is mine)

Ordering

This branch is cut from main, so on its own CI still hits the 2 deepeval collection errors that #609 fixes — verified with a clean uv sync --dev:

ERROR tests/.../third_party/deepeval/test_adapter.py
ERROR tests/.../third_party/deepeval/test_error_handling.py
Interrupted: 2 errors during collection

Kept as a draft for that reason. #609 then this turns ci.yml green; merge order the other way leaves 4 failures. Happy to rebase on #609 if you'd prefer to see this one green before review.

Related: #608 (integration workflow), #609 (dev-group pin).

[github-actions]

✅ No Breaking Changes Detected

No public API breaking changes found in this PR.

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)

[jariy17]

Draft. Depends on #609 — see Ordering below.

Fixes the 4 pre-existing unit-test failures on main that #609 does not cover. All four are tests (or a message/doc) that fell behind source and upstream changes; none are product bugs.

1 + 2. TestAsyncMode multi-agent and bidi callbacks

test_agentcore_memory_session_manager.py:3726 and :3756 reached into the private registry dict:

callbacks=registry._registered_callbacks.get(event_type, [])
assertall(asyncio.iscoroutinefunction(cb) forcbincallbacks) # always False

Upstream Strands now stores callbacks wrapped in a _CallbackEntry:

type= <class 'strands.hooks.registry._CallbackEntry'>
repr= _CallbackEntry(callback=<function ...register_hooks.<locals>._offload.<locals>._callback ...>)

iscoroutinefunction() on the wrapper is always False, so the assertion could never hold. The source is correct — session_manager.py:989-991 does register async _offload callbacks.

The fix is the accessor the passing sibling tests already use (lines 3636–3714): public registry.get_callbacks_for(event), which unwraps.

Side benefit: the bidi-init assertion at :3751 is not any(iscoroutinefunction(...)). It was passing for the wrong reason — the wrapper is never a coroutine function, so it would have passed even if the callback were wrongly async. Against unwrapped callbacks it now actually tests what it claims.

3. EvaluatorOutput validator message — source fix

The message advertised a value escape hatch that does not exist:

ifnotself.errorCodeandself.labelisNone: # value is never inspectedraiseValueError("Either label, value, or errorCode must be set; ...")

The class docstring (models.py:63-65) and every caller in src/ agree that label is required unless errorCode is set, so the message was the defect, not the logic. Corrected to "label is required for success responses; ...", which is also what the test expected.

4. test_evaluation_with_empty_trajectory

TypeError: 'EvaluationReport' object is not subscriptable

run_evaluations() returns a single report now, verified against the installed library:

run_evaluations -> <class 'strands_evals.types.evaluation_report.EvaluationReport'>
model_fields: ['overall_score', 'scores', 'cases', 'test_passes', ...]

Dropped the [0]. Also fixed the same stale pattern in README.md (2 occurrences) — the documented reports = ...; report = reports[0] snippet would hand users the identical TypeError.

Testing

• uv run pytest tests/ on Python 3.10 → 2934 passed, 10 skipped, 4 xpassed, 0 failed
• TestAsyncMode in isolation → 8 passed
• ruff check src/ tests/ → all checks passed; ruff format applied (the shortened message now fits one line — verified main was format-clean beforehand, so that reflow is mine)

Ordering

This branch is cut from main, so on its own CI still hits the 2 deepeval collection errors that #609 fixes — verified with a clean uv sync --dev:

ERROR tests/.../third_party/deepeval/test_adapter.py
ERROR tests/.../third_party/deepeval/test_error_handling.py
Interrupted: 2 errors during collection

Kept as a draft for that reason. #609 then this turns ci.yml green; merge order the other way leaves 4 failures. Happy to rebase on #609 if you'd prefer to see this one green before review.

Related: #608 (integration workflow), #609 (dev-group pin).

[github-actions]

✅ No Breaking Changes Detected

No public API breaking changes found in this PR.

[agentcore-devx-automation]

Claude Security Review: no high-confidence findings. (run)