👋 👋
Severity: Critical
File: confidence/confidence.py:80

I forked the repository and made a branch but wanted to respect the contribution guidelines before I opened a pull request.

description

context is declared as a class-level attribute with a mutable default dict:

class Confidence:
     context: Dict[str, FieldType] = {}

because__init__never assigns self.context = {}, all instances share the same dict object at the class level. Any call to put_context on one instance mutates that shared dict, silently contaminating every other Confidence instance in the process.

impact

• context set on one Confidence instance leaks into all other instances.
• child instances created via with_context are safe (they assign a new dict), but the root instance and any sibling roots
  are not.
• in a multi-tenant or multi-user scenario, this is a data isolation failure.

reproduction

from confidence.confidence import Confidence

a = Confidence("secret-a")
b = Confidence("secret-b")

a.put_context("user", "alice")
print(b.context)  # {"user": "alice"} — wrong, b is contaminated

Bug confirmed. b was never touched, but it has Alice's context because both instances share the same class-level dict.

fix

1. context (critical)
   add self.context: Dict[str, FieldType] = {} inside __init__, and change the async_client default to None with a guard in the body.

2. async_client (secondary) Change the default from httpx.AsyncClient() (evaluated once at class definition time, shared across all instances) to None, instantiating a fresh httpx.AsyncClient() per instance in the body of __init__. This prevents shared client state under concurrent use.

3. Regression test added tests/test_confidence.py::TestConfidence::test_context_is_isolated_per_instance — creates two root instances, mutates one, and asserts the other is unaffected.