Testing Fuzzing

Use this skill when the work calls for fuzz tests, property tests, randomized scenario tests, model-based tests, or replayable failure reduction. The goal is not "more random tests"; it is to expose input classes and state transitions that example tests miss, then turn each useful failure into a deterministic regression.

Operating Rules

1. Start with the target boundary. Identify the parser, serializer, validator, protocol handler, state machine, scheduler, allocator, CLI surface, API contract, or workflow being stressed.
2. State the risk. Name the defect class the randomized test is meant to find: panic, crash, data loss, invalid state transition, auth bypass, resource blow-up, nontermination, incorrect round trip, incompatible decode, or divergent implementation behavior.
3. Define the oracle before the generator. Random input without a verdict function creates noise.
4. Record every random seed, corpus path, command, timeout, and environment detail needed to reproduce a failure.
5. Minimize failures before fixing them. Keep the minimized case as a regression test and, when useful, as a seed corpus entry.

Oracle Selection

Choose one or more oracles that can be evaluated automatically: