rsun-diagnose

Follow a five-step discipline: first reproduce the bug with a runnable command, then minimise the failing input, instrument the code to observe real state, form a hypothesis from that data and write a minimal fix, and finally verify by re-running the original reproduction.

Create a command, curl request, or test that fails 100 percent of the time, such as running the failing script directly or calling the endpoint with a minimal payload. If you cannot reproduce it, stop and gather environment and timing details from the reporter first.

A minimal reproduction strips the failing scenario down to under 20 lines, removing unrelated fields and setup. This isolates the actual trigger, and the minimisation process itself often reveals the root cause before any instrumentation is needed.

Skip it for pure feature additions, which belong in a test-driven development flow, and for refactors where no bug exists. The workflow is designed specifically for defects: failing tests, broken endpoints, and malfunctioning scheduled jobs.

Without observing actual internal state such as response headers, variable values, or logs, hypotheses are guesses that often target the wrong layer. Instrumentation with print statements, verbose curl output, or log inspection confirms the root cause before any code changes.