pypi / monocle-apptrace

monocle-apptrace

What changed in monocle-apptrace 0.8, from its official release notes, upgrade guides and advisories. Sources checked 2026-10-19.

4 breaking1 security21 new APIs and options4 behavior and defaults1 tooling10 fixes and other

See which of these apply to your code, ranked, in your agent:

npx @unstale/cli init

@unstale/cli 0.2.0 on npm

monocle-apptrace 0.841 changes

Breaking changes 4

MS Agent Framework agents hosted on Azure AI Foundry now produce complete traces, identify Foundry as their host, and have spans retrievable from Okahu.
0.8.15release notes
setup_test_cases refuses discovery windows larger than this limit instead of silently truncating them; the value comes from the argument, then OKAHU_MAX_FACTS, then defaults to 1000.
0.8.15release notes
Turn spans now record prompt text using the renamed contents field instead of an object representation; the fallback also avoids exporting object reprs.
0.8.15release notes
AWS Bedrock AgentCore trace return now authorizes callers before returning spans; deployments must configure this key or a retrieval callback, and invocations must present the matching retrieval key.
0.8.13release notes

Security 1

Sensitive data in data.input and data.output is scrubbed before spans reach exporters; obfuscation is on by default and covers common credentials and secrets.
0.8.14release notes

New APIs and options 21

Accepts a testcase for parametrized tests.
0.8.15release notes
Accepts a testcase for parametrized tests.
0.8.15release notes
Accepts a testcase; every eval a case names is graded and all failures are reported together.
0.8.15release notes
Borrows the expected result from a different eval.
0.8.15release notes
Adds a runner for invoking an agent deployed to Azure AI Foundry and retrieving its spans from Okahu; it authenticates with an Entra bearer token and correlates spans using Foundry's trace id.
0.8.15release notes
Input and output assertions accept a testcase for parametrized tests.
0.8.15release notes
The root path / can now be opted into as a health check route and matches the root exactly; it is not a default.
0.8.15release notes
Remote trace retrieval now waits up to this many seconds for spans to reach the backend; the default is 60 seconds instead of a hard-coded 10.
0.8.15release notes
Sets the Okahu API timeout, which defaults to 120 seconds; precedence is explicit argument, environment variable, then default, and unusable values are logged and ignored.
0.8.15release notes
Recorded populations can become parametrizable test cases for eval tuning and A/B testing; parametrized cases are named after the case rather than [testcase0].
0.8.15release notes
Accepts a testcase for parametrized tests.
0.8.15release notes
Accepts a testcase for parametrized tests.
0.8.15release notes
Builds one FluentTestCase per recorded Okahu fact, or loads a committed JSON array with source="local" for network-free golden datasets.
0.8.15release notes
Provides the workflow name needed to retrieve Foundry runner spans; alternatively, use FOUNDRY_TRACE_WORKFLOW.
0.8.15release notes
Accepts a testcase for parametrized tests; trace sources also accept a start_time/end_time window.
0.8.15release notes
Setting this environment variable to true disables span obfuscation.
0.8.14release notes
Use this environment variable to scope span obfuscation to selected span types.
0.8.14release notes
Configure span obfuscators, including adding PII detection with credentials,presidio or disabling obfuscation with none.
0.8.14release notes
Pass an empty span_obfuscators list to disable span obfuscation.
0.8.14release notes
Implement a custom SpanObfuscator and configure it by its module:ClassName path.
0.8.14release notes
Health-check routes can be replaced via this variable; matching response-body checks are sampled down to one span per reset counter, while requests with params or a body and failing checks are exported.
0.8.13release notes

Behavior and default changes 4

A fact whose spans fail to load is emitted as a failing test under its own id, with the reason available in load_error.
0.8.15release notes
The LiteLLM metamodel now recognizes ReAct-style text tool calls and extracts their tool name.
0.8.14release notes
The LiteLLM metamodel now recognizes ReAct-style text tool calls and extracts their tool type.
0.8.14release notes
eval_name accepts a built-in eval template name or a custom template JSON file path, so callers need not switch to template_path for custom templates.
0.8.13release notes release notes

Tooling 1

The test generator discovers evals configured for an agent and adds matching check_eval assertions, supporting custom templates and searching eval traces beyond 90 days.
0.8.13release notes

Fixes and other changes 10

Streaming inference now emits a span when the consumer stops reading early; closing after full consumption cannot emit a duplicate span.
0.8.15release notes
A path-like string eval name is now classified as a custom template, matching check_eval, rather than being requested as a built-in template.
0.8.15release notes
The runner passes memory= only to callables that accept it and binds it with functools.partial for run_in_executor, which takes no kwargs.
0.8.15release notes
An unset API key is no longer stripped unguarded.
0.8.15release notes
An endpoint environment variable that is set but empty now falls back to the default endpoint instead of producing a scheme-less URL.
0.8.15release notes
Okahu enumerators now walk every page, and page size is bounds-checked locally instead of producing an opaque HTTP 400.
0.8.15release notes
Wrappers no longer stop the scope token twice, preventing a cleanup error from masking the test's original exception.
0.8.15release notes
Expected-failure marks now apply correctly to fluent assertions, avoiding hard failures for intentional failures and passes when a test silently stops failing.
0.8.15release notes
Resetting span processors now re-appends the trace-return processor, so trace return is not silently dropped on an initialized instrumentor.
0.8.15release notes
Span obfuscation preserves sentence punctuation after an unquoted credential value while still treating dots inside the value as part of the secret.
0.8.14release notes