forge

This is typically order-dependent shared state. Lock the test order with a fixed seed, isolate the minimal pair of tests that triggers the failure, then check for module-level globals, database rows, or temp files mutated by an earlier test. Reset that state in setup to confirm the cause.

Trace the code as the compiler or interpreter would: verify every import exists, check that each library method exists in the installed version, confirm types line up, and ensure every variable is defined before use. AI-generated code often invents plausible but nonexistent API methods.

It defines explicit standards for Python (type hints, pathlib, dataclasses), TypeScript (strict typing, async/await), Go (error wrapping, table-driven tests), Rust (Result propagation, no unwrap), and SQL (parameterized queries, no SELECT *).

Compilation only checks syntax, not semantics. Hallucinated library methods are often syntactically valid attribute lookups that only fail at runtime. Verify each method against the installed package version's documentation before trusting the code.

Ask one specific question only when the ambiguity changes the architecture. When a reasonable default exists, state the assumption explicitly and deliver the complete solution rather than stalling on clarification.