Read the user request and extract the symptom, expected behavior, actual behavior, environment, and constraints.

Inspect repository state:

git status --short
git branch --show-current
git log --oneline -10

Locate relevant code with fast search:
```
rg "<important term>"
rg --files
```
Read the smallest useful set of files before forming a conclusion.
Reproduce when feasible with existing tests, scripts, or a focused manual command.
Trace the data/control flow from entry point to failing behavior.
Check whether this may be a regression:
```
git log --oneline -- <relevant-file>
git blame <relevant-file>
```
Use history only when it helps explain timing or prior behavior.
Identify likely root cause, affected files, constraints, edge cases, risks, and missing information.
If multiple fixes are plausible, compare tradeoffs and recommend one.
Do not edit files unless the user explicitly asks to continue into implementation.

Output

Return:

Current understanding of the issue.
Evidence with file paths, line references, command results, or observed behavior.
Most likely root cause, clearly labeled if it is still an inference.
Proposed fix strategy.
Files likely to modify.
Tradeoffs when multiple solutions exist.
Tests or checks that should validate the fix.
Open questions only when the answer cannot be discovered locally.

Guardrails

Prefer concrete evidence over guesses.
Call out uncertainty directly.
Avoid broad refactors in the proposed fix unless the evidence shows they are necessary.
Keep the final investigation short enough to act on.