Skill systematic-debugging

Hard rule: do not package a guess as a permanent fix before locating the root cause. If the user explicitly needs emergency containment, you may apply a reversible temporary mitigation first, but you must state its risks and that the root cause is not fixed, then continue investigating.

1. Reproduce

• Read the error message and full stack trace carefully—the answer is often there. Do not skip them. Note the line number, file, and error code.
• Establish a reliable reproduction first: which steps trigger it every time? If you cannot reproduce it reliably, gather data before proceeding. Do not guess.
• Inspect recent changes: git diff, recent commits, new dependencies, and configuration or environment differences.

2. Trace backward to the root cause

• Follow the bad value backward: where did it come from → who called something with it → continue to the source. Fix it at the source, not at the symptom.
• In a multi-component system (CI→build→signing, API→service→library), record what enters and leaves each boundary. Run it once to see which layer fails, then investigate that layer. Do not guess based on intuition.
• Find a similar working example and list every difference from the broken one. Do not assume that "this difference cannot matter."
• Example: an amount on a page has two extra decimal places. Trace it backward: the display layer receives the wrong value → it came from the API → the value read from the database is correct → the API returned cents as dollars. Fix the API layer; do not compensate with display formatting.

3. One hypothesis at a time

• Write it down: "I believe the root cause is X because Y."
• Test it with the smallest change, varying only one thing at a time. Do not stack new fixes on an unverified one.
• If you do not understand X, say so. Do not pretend and guess.

4. Fix the root + add a regression test

• Write a failing test that reproduces the bug before fixing it (see tdd).
• Make one fix in one place. Do not opportunistically refactor or change anything else.
• Verify that the test passes, nothing else broke, and the problem is actually gone (see verify-before-done).

Three failed fixes = stop and question the architecture

If every fix creates a new problem elsewhere or requires a "large refactor," the problem is not the hypothesis but the architecture. Do not attempt a fourth fix. Stop and discuss whether the design itself is fundamentally wrong.

No random attempts: when these thoughts appear, stop and return to Step 1

Thought	What to do
"Apply a quick fix now and investigate the root cause later."	There is no later—once the symptom changes, you will stop investigating. Trace it to the root cause before changing anything.
"Change this and see whether it works."	That is guessing. First write, "I believe the root cause is X because Y," then test it with the smallest change.
"It is probably X; just fix it."	Reaching a conclusion before tracing the data flow is a gamble. Follow the bad value to its source first.
"Put several changes in and run it."	Change only one variable at a time, or a passing result will not tell you which change mattered.
"Two attempts failed; maybe one more will work."	Stop. A third attempt probably means the direction is wrong. Return to Step 1 or ask someone else.