What problem does it solve? Implementation-aware tests tend to avoid patterns the code cannot handle, so edge cases in a code-to-block round-trip pipeline go undetected. This Skill uses two agents with strict information boundaries to write realistic programs and verify round-trip correctness, exposing bugs that hand-written tests miss. ## Core Features & Use Cases - Dual-agent information isolation: Agent A writes real programs knowing only the language spec, while Agent B validates round-trip correctness with full access to the Semorphe source code. - Multi-language support: Works with C++, Python, Java, and JavaScript, with per-language compile/run commands and difficulty calibration (easy, medium, hard) plus scope filters like loops, functions, or arrays. - Structured result classification: Categorizes outcomes as PASS, SEMANTIC_DIFF, COMPILE_FAIL, LIFT_FAIL, EXPECTED_DEGRADATION, SCAFFOLD_LEAK, ROUNDTRIP_DRIFT, WRONG_COMPONENT, or TIMEOUT, and converts every result into permanent Vitest regression tests. - Use Case: Run cpp hard loops 20 to generate 20 difficult C++ loop programs, lift each to a SemanticTree, regenerate code, compare program output, and file any semantic differences as bugs with reproduction tests. ## Quick Start Ask the AI to run component-fuzz with a target language, difficulty, scope, and program count, for example fuzzing twenty hard C++ loop programs through the round-trip pipeline.
