[P2] Format-sensitivity fuzz testing for equivalent schemas

[cursor]

Summary

Measure whether harmless schema formatting changes alter model routing behavior. Transformations must preserve semantic/schema meaning (property order, description whitespace, equivalent JSON Schema representations).

Priority

priority:p2 — Research + robustness; command sketch: tool-semantics fuzz-format snapshot.json.

Suggested labels

enhancement, priority:p2, area:probes, research, area:cli

Acceptance criteria

• Generate multiple equivalent schema variants from a snapshot
• Compare baseline vs perturbed selection/accuracy with fake or opt-in model runner
• Emit FORMAT SENSITIVITY WARNING when delta exceeds threshold
• Transformations documented as meaning-preserving
• CI uses fakes only

Related

• Progressive tool-catalog discovery regression testing #86 discovery, Tool-description sensitivity and catalog-scaling research harness #95 description sensitivity, Detect tool collisions and emit confusability scores #79 collisions