Same-Topic Reinforcement and Conflict

Multiple instructions on the same topic must agree in direction. Conflicting instructions on the same topic destroy compliance catastrophically -- the model cannot follow both and may follow neither.

Antipatterns

• Writing "Use ruff for formatting" in one file and "Use black for formatting" in another. Same topic, opposite directives -- the model picks one unpredictably.
• Adding nuanced exceptions without scoping them: "ALWAYS use pytest" alongside "Don't use pytest for integration tests" reads as a contradiction without explicit conditional scoping.
• Restating a directive with weaker language elsewhere. "NEVER push to main" in one file and "Avoid pushing to main" in another creates ambiguity about whether the constraint is absolute.

Pass / Fail

Pass

<!-- file: .claude/rules/testing.md -->
Use `pytest` for all test files in `tests/`.
Run `uv run pytest tests/ -v` before committing.

Fail

<!-- file: .claude/rules/testing.md -->
Use `pytest` for all tests.
<!-- file: .claude/rules/workflow.md -->
Use `unittest` for all tests.

Fix

Remove or resolve conflicts first. Then check for weak reinforcement: strengthen the weak instruction (name constructs, use imperative modality) or remove it. Reinforce only with instructions of comparable strength.

Limitations

Detects same-topic instruction pairs using embedding similarity and opposite direction. May flag intentional nuance (e.g., "prefer X" with "but use Y when Z") as a conflict when the instructions are meant to coexist with different scopes.