Systematic Debugging

Use this skill to stop guessing, gather evidence, and narrow the actual cause before attempting a fix.

1. Core Defaults

• Reproduce the problem before changing code.
• Read the exact error, output, or observed behavior first.
• Change one variable at a time.
• Compare the broken path with a nearby working path when possible.
• Prefer the smallest probe that can confirm or kill one hypothesis.
• Stop after repeated failed fix attempts and re-check the mental model.

2. Investigation Loop

1. State the failure clearly.

   • observed behavior
   • expected behavior
   • where it happens

2. Capture the cheapest reliable reproducer.

   • failing command
   • failing test
   • exact log line
   • concrete input that triggers the problem

3. Read the relevant files in full.

   • do not patch based on a snippet alone
   • include nearby config or wrapper files when they shape the behavior

4. Check recent change surfaces.

   • git diff
   • recent commits
   • changed config
   • dependency or environment shifts

5. Compare with a working pattern.

   • sibling file
   • older version
   • another command path
   • similar code path that still behaves correctly

6. Form one hypothesis.

   • say what should be true if the hypothesis is correct
   • run the narrowest probe that can prove or disprove it

7. Only after the cause is credible, hand off to execution.

   • use local tdd-tidy-first for the smallest verified code or config change

3. Repo Fit

• Prefer rg for text and file discovery.
• Record the reproducer command and the exact output you observed.
• Use mkmd research artifacts when the debugging trail will span many steps or multiple turns.
• Prefer POSIX-compatible and repo-managed tools over ad hoc global setup.
• Do not hide uncertainty with speculative cleanup or defensive code.

4. Stop Conditions

Stop and reassess when any of these happen:

• three fix attempts failed
• the reproducer changed unexpectedly
• the observed behavior contradicts the current hypothesis
• the failure depends on permissions, hooks, or environment boundaries outside the current lane

At that point, summarize:

• what is proven
• what is still unknown
• which hypothesis failed
• what to test next

5. Handoff To Implementation

When the likely cause is clear, switch to tdd-tidy-first and keep the next step narrow:

• add or tighten the reproducer when cheap
• make the smallest change that should fix the proven cause
• run the fastest relevant verifier
• widen verification only after the narrow slice passes