python-code-audit-sweep

Python code-audit sweep

When to use

The user says something like "check the code for improvements, no feature changes" or "do a general cleanup pass". You have a Python codebase, no specific bug report, and need to surface high-signal, low-risk findings without accidentally changing behavior.

Problem

A naive "cleanup" PR usually fails review because it mixes three very different kinds of change into one blob:

• A real latent bug (semantics change when fixed).
• Dead code / unused locals (safe, mechanical).
• Spelling, wording, anti-patterns in comments (pure style).

Reviewers want to merge (2) and (3) fast and scrutinize (1) carefully. If you stuff them into one PR, the whole thing blocks on the hardest item, and the diff hides the real bug behind dozens of typo fixes.

Solution

Do the audit in two phases: find, then partition into separate PRs.

1. Find (cheap, parallel grep + pyflakes)

# Static analysis: unused imports/vars, undefined names, simple smells.
pyflakes src/ | tee /tmp/pyflakes.txt

# Common anti-patterns (extend as needed).
grep -rnE 'not [a-zA-Z_][a-zA-Z_0-9.()\[\]]* is None' src/   # should be "is not None"
grep -rnE '== None|!= None' src/                             # should be "is (not) None"
grep -rn  'type(str)\|type(int)\|type(list)\|type(dict)' src/ # type() applied to a builtin type

# Typo shortlist — curated, not exhaustive; avoids false positives.
grep -rniE '\b(seperat|writen|recieve|occured|begining|lenght|tranform|initialis|standardalone|dependancy|mutiple|visibilty|accomodate|seperately)\w*' \
    src/ doc/ tool/

# Filenames with unusual punctuation (real bugs on GitHub's Markdown renderer).
find . -name '*,md' -o -name '*.m d' -o -name '*.mdd'

2. Partition into three PRs

Label every finding as A, B, or C. One PR per label, in this order:

Tag	Contents	Risk	Review burden
A	Real bugs: broken error messages, dead-assignment-that-hides-a-branch, wrong file extensions.	Behavior may change (even if tiny).	Highest — maintainer must read carefully.
B	Dead locals, unused imports, tuple-unpacking where half is unused.	None.	Low.
C	Typos in comments/docs, not x is None → x is not None, file renames with no in-repo references.	None.	Lowest.