Code Review

When to Use This Skill

• •User asks for a code review of recent changes.
• •User wants feedback on code quality, correctness, or style.
• •User is preparing a PR and wants a pre-review check.

Goal

Review code changes with a focus on:

1. •Correctness: Does the code do what it's supposed to?
2. •Edge cases: Are error conditions handled?
3. •Style: Does it follow project conventions?
4. •Performance: Are there obvious inefficiencies?

Workflow (Agent Instructions)

Prefer reviewing the actual diff (what changed) rather than only reading current file contents.

1 Determine review scope

1. •Identify what you are reviewing:

• •Local working tree changes, or
• •A branch/PR compared to a base branch.

1. •For local changes, start with:

• •git status --porcelain

1. •For branch/PR-style review, pick a base branch (try main then master) and use:

• •git diff <base>...HEAD --stat
• •git diff <base>...HEAD

2 Summarize what changed

1. •Run:

• •git diff --stat
• •git diff --cached --stat

1. •Provide a 3–6 bullet summary of the intent of the changes.

3 Review checklist (apply to the diff)

Correctness

• •Look for logic errors, incorrect assumptions, and mismatched units/types.
• •Validate function contracts: inputs/outputs, return types, side effects.
• •Check that refactors preserve behavior (renames, moved code, split functions).

Edge cases

• •Null/empty input handling.
• •File/path handling (prefer pathlib in this repo).
• •Time-series/data issues: missing timestamps, empty DataFrames, NaNs.
• •Defensive behavior: informative errors, safe fallbacks.

Style / Conventions

• •PEP 8 formatting and readable naming.
• •Prefer small, composable functions.
• •Type hints when practical.
• •Keep analysis logic aligned with existing conventions (7-day/30-day, 3-day comparisons).
• •For CLI output, prefer existing rich patterns.

Performance

• •Identify obvious pandas anti-patterns (row-wise .apply when vectorization works, repeated groupby/merge in loops).
• •Avoid repeated file reads; prefer loading once and passing DataFrames through.
• •Watch for N^2 operations on large datasets.

4 Scan for common foot-guns

Use diff + grep/search tools available in your shell/editor to flag:

• •breakpoint(, pdb, print( (unless clearly intentional)
• •TODO / FIXME
• •commented-out blocks added

Examples (PowerShell):

• •git diff | Select-String -Pattern "breakpoint\(|print\(|TODO|FIXME"
• •git diff --cached | Select-String -Pattern "breakpoint\(|print\(|TODO|FIXME"

5 Basic validation

1. •Run unit tests:

• •uv run pytest

1. •If tests fail, report the failing tests and key traceback lines.

How to Provide Feedback

• •Be specific about what needs to change.
• •Explain why, not just what.
• •Suggest alternatives when possible.
• •Distinguish blocking issues (bugs, crashes, wrong results) from non-blocking improvements (style, minor refactors).

Output Template

CODE REVIEW

Summary:
- ...

Blocking Issues:
- ...

Non-blocking Suggestions:
- ...

Risk / Notes:
- ...

Validation:
- Diff reviewed: (staged/unstaged or <base>...HEAD)
- Tests: PASS/FAIL (details)

Recommendation: APPROVE / REQUEST CHANGES