How to Debug Python#

Use gptme to reproduce an error, trace through it, and apply a fix — without leaving the terminal.

Fix a failing test#

Pass the test output and the relevant files:

python -m pytest tests/test_api.py -x 2>&1 | gptme 'fix the failing test' src/api.py tests/test_api.py

The -x flag stops at the first failure so the output is focused.

Pipe the error and the source file:

python myapp.py 2>&1 | gptme 'explain this traceback and fix the root cause' myapp.py

gptme 'users report that login fails when the username contains a space; reproduce the bug and fix it' src/auth.py tests/test_auth.py

gptme will write a failing test, run it, trace the error, and apply the fix.

python -m cProfile -s cumulative myapp.py 2>&1 | gptme 'identify the bottleneck and suggest a fix' myapp.py

gptme 'add temporary debug logging to trace the request path, run the failing test, then remove the logging' src/handler.py

gptme 'this test fails intermittently; find the source of flakiness and fix it' tests/test_cache.py src/cache.py

Useful pattern: tell gptme to run the test multiple times and look for patterns.

mypy src/ 2>&1 | gptme 'fix all type errors' src/

Run the test inside gptme: gptme 'run the tests and fix failures' lets it iterate without you copy-pasting output.
Include the full error: don’t truncate the traceback — the line numbers matter.
Chain the fix with a test: end with - 'run the tests again and confirm they pass' to close the loop.