Agent-Skills.md

Agent Skills: Review Code

Perform comprehensive multi-agent code review for PRs, commits, or entire

UncategorizedID: mgiovani/cc-arsenal/review-code

Repository

mgiovani/cc-arsenal
mgiovaniLicense: MIT
21

Install this agent skill to your local

pnpm dlx add-skill https://github.com/mgiovani/cc-arsenal/tree/HEAD/skills/review-code

Skill Files

Browse the full folder contents for review-code.

Download Skill

Loading file tree…

skills/review-code/SKILL.md

Skill Metadata

Name
review-code
Description
Runs a comprehensive multi-agent code review of a PR, commit, or the

Review Code

Cross-Platform AI Agent Skill This skill works with any AI agent platform that supports the skills.sh standard.

Code Review

Comprehensive multi-agent code review covering correctness, performance, code style, test coverage gaps, and error handling. This skill performs analysis only - it identifies issues, explains findings, and suggests improvements without making code changes.

Anti-Hallucination Guidelines

CRITICAL: Code reviews must be based on ACTUAL code analysis and VERIFIED patterns:

  1. Read before claiming - Never report issues in code that has not been read
  2. Evidence-based findings - Every finding must reference specific file paths and line numbers
  3. Pattern matching - Use Grep to find actual problematic patterns, not hypothetical ones
  4. Quantifiable results - Count actual instances, do not estimate
  5. No false positives - Verify each finding matches documented issue patterns
  6. Scope verification - Only review files within specified scope (PR/commit/all)
  7. Respect conventions - Understand project patterns before flagging style issues
  8. Context matters - A pattern acceptable in one context may be problematic in another

Review Workflow

Phase 0: Determine Review Scope

Parse arguments to determine what to review:

Arguments:
- <pr_number>: Review only files changed in PR (e.g., "123", "#123")
- <commit_sha>: Review only files changed in commit (e.g., "abc123")
- "--all" or no args: Review entire codebase
- "--focus [correctness|performance|style|tests|errors]": Focus on specific review dimension
If PR or commit specified, use Bash to get changed files and diff context:
```bash
# For PR - get files and full diff
gh pr view <pr_number> --json files --jq '.files[].path'
gh pr diff <pr_number>

# For commit
git diff-tree --no-commit-id --name-only -r <commit_sha>
git show <commit_sha>
**Important**: When reviewing a PR or commit, always retrieve the full diff. The diff context is essential for understanding what changed vs. what was already there. Agents should focus findings on **changed lines** while using surrounding code for context.

### Phase 1: Project Discovery

Explore the codebase to understand the project's technology stack, conventions, and quality standards:

### Phase 2: Initialize Progress Tracking

Use TodoWrite to track review progress across all specialist dimensions and report generation.

### Phase 3: Parallel Specialist Review

Spawn 5 parallel Explore agents for comprehensive code review. Each agent specializes in a specific review dimension. For detailed agent prompts and patterns, see [references/agent-prompts.md](references/agent-prompts.md).

**Agent assignments:**
- **Agent 1**: Correctness & Logic — bugs, race conditions, off-by-one errors, null safety, type mismatches
- **Agent 2**: Performance — algorithmic complexity, unnecessary allocations, N+1 queries, missing caching, memory leaks
- **Agent 3**: Code Style & Patterns — naming, structure, DRY violations, SOLID adherence, framework idioms
- **Agent 4**: Test Coverage Gaps — untested code paths, missing edge case tests, weak assertions, test quality
- **Agent 5**: Error Handling & Edge Cases — unhandled exceptions, missing validation, boundary conditions, graceful degradation

Each agent must:
1. Grep for issue patterns across files in scope
2. Read each match to verify context and confirm it is a genuine issue
3. Extract exact code snippets (5-10 lines) with file:line references
4. Explain why the code is problematic
5. Classify severity (Critical/Major/Minor/Nit)
6. Provide a concrete fix suggestion with code example

**Severity Definitions:**
- **Critical**: Bugs that cause data loss, crashes, security holes, or incorrect business logic
- **Major**: Significant issues affecting reliability, performance degradation, or maintainability risks
- **Minor**: Improvements for readability, consistency, or minor inefficiencies
- **Nit**: Style preferences, cosmetic suggestions, optional improvements

### Phase 4: Consolidate & Analyze Findings

After all agents complete:

1. **Collect all findings** from the 5 parallel agents
2. **Deduplicate** - Remove duplicate findings across agents (e.g., the same function flagged by both correctness and error handling agents)
3. **Prioritize by severity**:
 - **Critical**: Data corruption, crashes, security implications, broken business logic
 - **Major**: Performance bottlenecks, reliability issues, test gaps for critical paths
 - **Minor**: Code readability, minor inefficiencies, style inconsistencies
 - **Nit**: Naming preferences, optional simplifications, cosmetic changes
4. **Categorize by dimension**: Group findings under the 5 specialist categories
5. **Cross-reference**: Note findings that span multiple dimensions (e.g., a missing null check is both a correctness and error handling issue)
6. **Statistics**: Count total findings by severity, by dimension, files reviewed vs. files with issues

### Phase 5: Generate Review Report

Generate a comprehensive markdown report following the template in [references/report-template.md](references/report-template.md).

**Report sections:**
1. Executive summary with overall code quality assessment
2. Severity breakdown with counts
3. Findings organized by dimension, each with file:line, code snippet, explanation, and fix suggestion
4. Prioritized action items (Critical first, then Major)
5. Positive observations - highlight well-written code, good patterns, thorough tests

### Phase 6: Iterative Re-Review (Diff-Only Re-Scan)

**This is the key differentiator.** After the initial review, if the user makes fixes and requests a re-review:

1. **Detect changes since last review**:
 ```bash
 # Get files changed since the review started
 git diff --name-only HEAD@{<timestamp>}..HEAD
 # Or if on a branch with new commits
 git diff --name-only <last_reviewed_commit>..HEAD
 2. **Scope re-review to changed files only** - Do NOT re-scan the entire codebase
3. **Re-run only relevant agents** - If fixes were for performance issues, re-run Agent 2 (Performance) on the changed files
4. **Verify fixes** - Check that previously reported Critical/Major issues are actually resolved
5. **Report delta** - Show what was fixed, what remains, and any new issues introduced by the fixes

**Re-review output format:**
```markdown
## Re-Review Report (Diff-Only)

**Files re-scanned**: [N files changed since last review]
**Previous findings**: [N total]
**Resolved**: [N findings fixed]
**Remaining**: [N findings still present]
**New issues**: [N new findings from fixes]

### Resolved Findings
- ~~[Finding title]~~ — Fixed in `file.py:45`

### Remaining Findings
- [Finding title] — Still present in `file.py:30`

### New Findings
- [New finding from fix] — Introduced in `file.py:50`
To trigger a re-review, the user runs the skill again after making fixes. The skill detects that a review was recently performed (by checking git log for recent review-related commits or by the user explicitly stating "re-review") and automatically enters diff-only mode.

## Usage

```bash
# Review a specific PR
review-code 123
review-code #456

# Review a specific commit
review-code abc123def

# Review entire codebase
review-code --all
review-code

# Focus on a specific dimension
review-code 123 --focus performance
review-code --all --focus tests

# Re-review after fixes (run again on same PR)
review-code 123
## Focus Options

- `correctness`: Focus on bugs, logic errors, type safety, race conditions
- `performance`: Focus on algorithmic complexity, resource usage, caching, queries
- `style`: Focus on naming, structure, patterns, framework idioms, DRY/SOLID
- `tests`: Focus on test coverage gaps, assertion quality, edge case testing
- `errors`: Focus on error handling, validation, boundary conditions, graceful degradation

If no focus specified, perform comprehensive review across all dimensions.

## Additional Resources

- [references/agent-prompts.md](references/agent-prompts.md) - Detailed grep patterns and agent prompts for each review dimension
- [references/report-template.md](references/report-template.md) - Full markdown report template with all sections

## What This Skill Does

- Identifies bugs, logic errors, and correctness issues
- Analyzes performance bottlenecks and optimization opportunities
- Reviews code style, patterns, and architectural adherence
- Discovers test coverage gaps and weak test assertions
- Evaluates error handling, input validation, and edge cases
- Generates a comprehensive markdown report with actionable findings
- Supports iterative diff-only re-review after fixes

## What This Skill Does NOT Do

- Does not modify any code
- Does not automatically fix issues
- Does not commit changes
- Does not run tests or benchmarks
- Does not perform security-specific analysis (use review-security for that)
- Does not guarantee detection of all issues

## Limitations

- **Static analysis only**: Cannot detect runtime-only issues
- **Pattern-based**: May miss deeply context-specific problems
- **No dynamic testing**: Cannot measure actual performance impact
- **False positives possible**: Some findings may be intentional design choices
- **Requires manual review**: Expert judgment recommended for Critical findings
- **Language support**: Best coverage for Python, JavaScript/TypeScript, Go, Java; basic coverage for other languages

## Simplicity & Over-Engineering Lens (Claude Code enhancement)

LLM-written code tends to over-engineer: interfaces built for one implementation, factories for one product, wrapper layers that just forward a call. None of that shows up as a bug, so the five specialists in Phase 3 don't catch it — it needs its own lens. This module adds a 6th parallel specialist and a matching report dimension.

### Agent 6: Simplicity & Over-Engineering

Spawn this agent alongside Agents 1-5 in Phase 3, in the same parallel batch:

Agent 6 - Simplicity & Over-Engineering (Explore, Haiku): prompt: "Review [SCOPE] for unnecessary complexity — code that does more than the current, concrete requirement needs.

1. Grep for interfaces/abstract classes/protocols with exactly one concrete
   implementation, factories that construct exactly one product, wrapper
   functions or classes that only forward calls without adding behavior, and
   configuration flags or parameters that no caller ever varies.
2. Read each match plus surrounding context to confirm it's genuinely
   unnecessary, not a documented extension point for a real second caller or
   plugin contract.
3. Before flagging something as speculative, check whether equivalent behavior
   already exists — in this codebase (search first), the standard library, or
   the framework/platform in use.
4. Classify each confirmed finding with exactly one tag:
   - [delete] — dead/unused code with no live caller
   - [reuse] — equivalent logic already exists elsewhere in this codebase
   - [stdlib] — the standard library already covers it
   - [builtin] — the framework/platform already provides it
   - [unneeded] — speculative code with no current caller (unused flag,
     unexercised branch, extension point nobody extends)
   - [simplify] — a one-implementation interface/factory, or a pure-forwarding
     wrapper, that should be inlined or merged
5. If a finding is really a correctness bug, a security hole, or a performance
   problem rather than just unnecessary complexity, do NOT tag it here — note
   it separately as out-of-scope-for-OE so it can be routed to the
   Correctness, Performance, or Error Handling reviewer instead. Complexity
   must never be used to mask, or be mistaken for, a real bug.
6. For each finding report: file:line, the one tag, a one-sentence description
   of what's unnecessary and why, the suggested deletion or replacement, and
   the number of lines that change would remove.
7. Sum the lines-removed across findings you are confident about, from code you
   actually read — not estimated. Report that sum as a static count only. Do
   not state or imply runtime, token, bundle-size, or percentage savings: the
   simplified version was never built or run, so there is no measured baseline
   to compare against. If a real benchmark already exists in the codebase for
   the code in question, you may cite it — otherwise say nothing about
   performance impact."

subagent_type: "Explore" model: "haiku"


**Routing out of scope**: when Agent 6 flags something that Phase 4 consolidation determines is actually a correctness, security, or performance issue, move it into the matching dimension (`CL-`, `PF-`, or `EH-` prefix) instead of reporting it as an OE finding. An over-engineered function that also happens to be buggy is a bug first.

### Report Addendum

Add a sixth dimension to Phase 4/5 output, alongside the existing five:

- **Dimension**: Simplicity & Over-Engineering
- **Finding prefix**: `OE-`
- **Default severity**: Minor or Nit. Escalate to Major only when the complexity itself causes a reliability or maintainability failure (e.g., a forwarding wrapper that silently drops an error the caller needs) — not merely because it exists.
- **Per-finding fields**: severity, `file:line`, the single tag from the list above, description, suggested deletion/replacement, lines-removed.

Report the aggregate as its own line, separate from the per-finding lines-removed counts:

> **Estimated lines removable (static count, not a benchmark): ~N**

`N` is the sum of lines-removed across all `OE-` findings, counted from code actually read during this review. Never state or imply a percentage, runtime, token, or bundle-size saving next to this number — no leaner version was built or measured. If the review surfaces a real, previously-measured benchmark for the flagged code, cite that benchmark by its source instead of inventing a figure.