pre-pr-review

Pre-PR Review

Review the current branch as a thorough, senior code reviewer. The goal is to catch issues before they reach human reviewers — saving review cycles and preventing convention drift.

This skill is for changes made outside the pipeline. If you just ran /pipeline, Stage 5 already reviewed the code — you don't need this skill.

Setup

1. Detect the base branch — do not assume main:

   BASE=$(git symbolic-ref refs/remotes/origin/HEAD 2>/dev/null | sed 's@^refs/remotes/origin/@@')
   if [ -z "$BASE" ]; then
     git show-ref --verify --quiet refs/heads/main && BASE=main || BASE=master
   fi
   echo "Base branch: $BASE"
   

2. Load the project's standards (read whichever exist):

   • AGENTS.md (coding standards, platform rules)
   • skills/code-conventions/SKILL.md (project coding conventions)
   • docs/audit/03-compliance.md (known conventions and anti-patterns from audit)
   • docs/audit/00-project-context.md (project context, commands)
   • CONTRIBUTING.md

3. Get the scope of changes:

   git rev-parse --abbrev-ref HEAD
   git log --oneline $BASE..HEAD
   git diff --name-only $BASE...HEAD
   git diff $BASE...HEAD
   

4. Check for merge conflicts with the base branch:

   git merge-tree $(git merge-base HEAD $BASE) $BASE HEAD | head -20
   

   If there are conflicts, flag them immediately — they must be resolved before any review makes sense.

5. Read each changed file in full (not just the diff). A change that looks correct in isolation may be wrong in context — duplicate of something nearby, inconsistent with an adjacent function, or breaking a pattern in the same file.

If the diff is very large (20+ files), ask the user: "This is a large diff. Should I review everything, or focus on specific areas?"

Review Checklist

Work through each area. For each issue found, report: the file and line, what's wrong, why it matters, and the suggested fix.

Correctness

• Does the logic do what it's supposed to? Trace the main code paths.
• Are edge cases handled (empty inputs, nulls, error conditions)?
• Are there off-by-one errors, wrong comparisons, or missed conditions?