≈1.3k tokens on demand. Your agent reads this file only when SKILL.md points to it.
Code Review Anti-Patterns
Review process anti-patterns, behavioral anti-patterns, and countermeasures.
1. Process Anti-Patterns (AWS Well-Architected)
| # |
Pattern |
Symptoms |
Impact |
Countermeasure |
| 1 |
Infrequent Reviews |
Reviews skipped or rarely performed |
Missed error detection, isolated development |
Make reviews a mandatory pre-merge step |
| 2 |
Excessive Reviewers |
Too many reviewers assigned |
Bottleneck, unnecessary delays |
Set practical reviewer count based on code complexity |
| 3 |
No Automated Feedback |
No automated tooling used |
Extended review time, focus on trivial issues |
Run SAST/Linters beforehand; humans focus on higher-order issues |
| 4 |
Large Batch Reviews |
Multiple changes reviewed at once |
Difficult to isolate issues, long review cycles |
Submit small, focused PRs |
| 5 |
Unconstructive Reviews |
Harsh tone, vague feedback |
Lowered developer morale, blocked dialogue |
Train on specific, constructive feedback |
| 6 |
No Action on Findings |
Findings left unaddressed |
Same issues recur, reviews become perfunctory |
Introduce feedback tracking systems |
2. Behavioral Anti-Patterns
8 Major Behavioral Patterns
| Pattern |
Description |
Judge's Countermeasure |
| Rubber Stamping |
Approving without reading content |
Automated review ensures minimum quality baseline |
| Nit-Picking |
Fixating on style/formatting trivia |
Delegate style to Linter/Zen; focus on logic |
| Long-Running PRs |
PRs stalled in review queue for extended periods |
Measure and alert on review turnaround time |
| Self-Merging |
Approving and merging one's own code |
Require at least one other reviewer |
| Heroing |
One person handles all reviews |
Distribute review load via rotation |
| Over-Helping |
Reviewer rewrites the code |
Limit to suggestions; let the author make fixes |
| "Just One More Thing" |
Repeated additional requests just before approval |
Clarify scope; defer additions to follow-up issues |
| Knowledge Silos |
Only specific individuals review specific areas |
Promote cross-reviews and documentation |
3. Cognitive Biases and Review Quality
Biases That Affect Reviews
| Bias |
Impact |
Countermeasure |
| Confirmation Bias |
Only detects issues matching existing beliefs |
Checklist-based reviews |
| Decision Fatigue |
Quality degrades in later portions of review |
Keep sessions under 60 minutes; take breaks |
| Anchoring |
First issue found biases subsequent evaluation |
Structured review order (Security → Logic → Performance) |
| Bandwagon Effect |
Conforming to other reviewers' opinions |
Independent review first, then discussion |
| Halo Effect |
Review depth varies based on author's reputation |
Evaluate code only; disregard author identity |
Judge's Cognitive Bias Countermeasures
Judge's automated review is unaffected by cognitive biases:
✅ Reviews all files at consistent depth
✅ Processes large volumes of code without fatigue
✅ Checklist-based approach prevents omissions
✅ Applies the same standards regardless of author
However, AI review has limitations:
❌ Incomplete understanding of business context
❌ Architecture decisions require human judgment
❌ Difficult to make judgments based on tacit knowledge
4. Review Scope Anti-Patterns
"Everything Everywhere All at Once" Pattern
❌ Checking all aspects in a single review:
Correctness + Security + Performance + Style + Tests + Documentation
✅ Separate concerns across reviews (Specialist-Agent pattern):
Pass 1: Correctness & Logic (Judge)
Pass 2: Security (Sentinel)
Pass 3: Performance (Bolt)
Pass 4: Test Quality (Radar)
Inconsistent Feedback
❌ Rejecting patterns that were approved last time
❌ Different standards across reviewers
❌ Unwritten implicit rules
✅ Codify review guidelines
✅ Living Rules (continuously updated team standards)
✅ AI review ensures standard consistency
5. Judge's Anti-Pattern Avoidance
| Anti-Pattern |
Judge's Countermeasure |
| Style nit-picks |
severity=INFO, delegate to Zen |
| Findings buried in noise |
Display CRITICAL/HIGH at top |
| No follow-up |
Explicitly name remediation agent |
| Insufficient context |
Intent-check via PR description + commit message |
| Excessive false positives |
Apply filtering from codex-integration.md |
Source: AWS: Anti-patterns for Code Review · CodeRabbit: 5 Code Review Anti-Patterns · IEEE: Anti-patterns in Modern Code Review · Arxiv: Towards Debiasing Code Review
