Agent skill

review

Code review with confidence-based filtering. Use to review code for bugs, security issues, and quality problems. Only reports issues with high confidence to reduce noise.

Stars 163
Forks 31

Install this agent skill to your Project

npx add-skill https://github.com/majiayu000/claude-skill-registry/tree/main/skills/other/review-paxtone-io-openkodo

SKILL.md

Code Review with Confidence Filtering

Overview

Review code thoroughly but only report issues you're confident about. Filter out noise - stylistic preferences and uncertain concerns don't belong in reviews. Focus on bugs, security issues, and clear violations.

Core principle: Only report issues with >= 80% confidence.

Announce at start: "I'm using the review skill to review this code."

When to Request Review

Mandatory:

  • After completing a feature
  • Before merging to main
  • After fixing complex bugs

Optional but valuable:

  • When stuck (fresh perspective helps)
  • Before major refactoring
  • After implementing unfamiliar patterns

The Review Process

Step 1: Gather Context

Get the diff:

bash
BASE_SHA=$(git merge-base HEAD main)  # or specific commit
HEAD_SHA=$(git rev-parse HEAD)
git diff $BASE_SHA..$HEAD_SHA

Check existing patterns:

bash
kodo query "code style"      # Project conventions
kodo query "error handling"  # Error patterns used

Step 2: Review for Issues

Priority categories:

Priority Confidence Action Examples
Critical >= 90% Must fix before merge Security holes, data loss, crashes
Important >= 80% Should fix Logic bugs, missing error handling
Minor >= 80% Nice to fix Performance, readability

DO report (>= 80% confidence):

  • Clear bugs with specific line numbers
  • Obvious security issues
  • Missing error handling that will cause failures
  • Logic errors you can prove with example input

DO NOT report (< 80% confidence):

  • Stylistic preferences ("I would have done X")
  • Uncertain concerns ("This might cause issues")
  • Opinions without evidence
  • Things that "feel wrong" but you can't explain why

Step 3: Document Findings

Report format:

markdown
## Code Review: [Feature/PR Name]

**Commits reviewed:** `BASE_SHA..HEAD_SHA`
**Files changed:** X files, +Y/-Z lines

### Strengths
- [What's done well - acknowledge good work]

### Issues

#### Critical: [Title] (Confidence: XX%)
**File:** `path/to/file.rs:42`
**Problem:** [Specific description]
**Evidence:** [Why you're confident this is wrong]
**Fix:**
```rust
// Suggested fix

Important: [Title] (Confidence: XX%)

...

Assessment

[ ] APPROVE - Ready to merge [ ] REQUEST CHANGES - Fix critical/important issues first


### Step 4: Handle Feedback

**If reviewer feedback is disputed:**
- Push back with technical reasoning
- Show code or tests that prove it works
- Ask for clarification on vague feedback

**If you're the reviewer and author disagrees:**
- Listen to their reasoning
- Verify your concern is valid (>= 80% confidence)
- Accept that you might be wrong

## Integration with Kodo

**Capture review insights:**
```bash
kodo reflect --signal "Found common bug pattern: missing null check in X"
kodo reflect --signal "Project uses Y pattern for error handling"

Check if similar issues were caught before:

bash
kodo query "review bugs"     # Past review findings
kodo query "common mistakes" # Known pitfalls

Confidence Calibration

90%+ confidence means:

  • You can explain exactly why it's wrong
  • You can show an input that triggers the bug
  • The issue violates documented requirements

80%+ confidence means:

  • You're fairly certain but can't prove it
  • The code contradicts established patterns
  • The behavior is clearly unintended

< 80% means:

  • Don't report it
  • Or preface with "Question:" not "Issue:"

Key Principles

  • High confidence only - Filter noise, report signal
  • Be specific - Line numbers, exact problems, concrete fixes
  • Acknowledge strengths - Reviews aren't just criticism
  • Push back respectfully - If reviewer is wrong, say so with evidence
  • Learn from reviews - Capture patterns with kodo reflect

Red Flags

You're doing it wrong if:

  • Reporting "I would have done it differently"
  • No line numbers or specific locations
  • Confidence below 80% on reported issues
  • Only criticism, no acknowledgment of strengths
  • Accepting invalid feedback without pushback
  • Not capturing insights for future sessions

Expand your agent's capabilities with these related and highly-rated skills.

Didn't find tool you were looking for?

Be as detailed as possible for better results