code-reviewer

Purpose

[TODO: Add purpose description]

MANDATORY WORKFLOW (MUST FOLLOW EXACTLY)

⚠️ STEP 1: Identify Review Scope (REQUIRED)

YOU MUST:

1. Invoke the get-git-diff skill to identify changed files, OR analyze user-specified files directly
2. Determine comparison scope:
   • Which commits/branches to compare? (e.g., HEAD^ vs HEAD, main vs feature-branch)
   • If not specified, default to comparing current changes against the default branch
3. Classify changed files by language and framework:
   • Detect languages from file extensions (.py, .js, .ts, .go, .java, .cs, .rb, etc.)
   • Identify frameworks from imports, config files, and project structure
4. If no reviewable files were changed, inform the user and exit gracefully
5. Focus subsequent review ONLY on the files identified in the diff

DO NOT PROCEED WITHOUT IDENTIFYING REVIEW SCOPE

⚠️ STEP 2: Load Memory & Context (REQUIRED)

YOU MUST:

1. CHECK PROJECT MEMORY FIRST:
   • Identify the project name from the repository root or ask the user
   • Use memoryStore.getSkillMemory("code-reviewer", "{project-name}") to load project-specific patterns
   • Cross-skill discovery: Use memoryStore.getByProject("{project-name}") to check for related skill results (e.g., python-code-review, dotnet-code-review, database-schema-analysis)
   • If memory exists: Review previously learned patterns, conventions, and project-specific context
   • If no memory exists (empty result): Note this is first review, you will create memory later
2. USE CONTEXT INDEXES FOR EFFICIENT LOADING:
   • For each detected language, use contextProvider.getDomainIndex("{language}") to understand available context
   • Use contextProvider.getDomainIndex("security") for security-related context
   • Use contextProvider.getCrossDomainContext("{language}", triggers) for cross-cutting concerns

See ContextProvider and MemoryStore interfaces.

DO NOT PROCEED WITHOUT COMPLETING THIS STEP

⚠️ STEP 3: Load Relevant Context (REQUIRED)

YOU MUST use the indexes to load only relevant files:

Based on languages and frameworks detected in Step 1:

1. ALWAYS: Use contextProvider.getAlwaysLoadFiles("{language}") for each detected language to load universal patterns and common issues
2. Based on framework detected: Use contextProvider.getConditionalContext("{language}", detection) to load framework-specific patterns
3. For security-sensitive code: Use contextProvider.getCrossDomainContext("{language}", triggers) where triggers include:
   • Authentication/authorization code
   • User input handling and validation
   • Database queries and data access
   • File system operations
   • Network/API calls
   • Cryptographic operations
4. For multi-language PRs: Load context for EACH language present in the changeset, respecting the file token budget

Progressive loading: Only load files relevant to the detected languages, frameworks, and code type. The ContextProvider respects the 4-6 file token budget automatically.

DO NOT SKIP LOADING RELEVANT CONTEXT FILES

⚠️ STEP 4: Deep Review (REQUIRED)

YOU MUST examine ONLY the changed code for ALL categories below:

Important: While reviewing changed lines, consider the surrounding context to understand:

• How changes interact with existing code
• Whether changes introduce regressions
• Impact on callers and dependent code
• Whether the change addresses the root cause or masks symptoms

Review Categories (language-agnostic):

Correctness & Logic: Algorithm correctness, boundary conditions, null safety, type correctness, edge cases, off-by-one errors, integer overflow, state corruption
Error Handling: Missing error handling, swallowed exceptions, improper error propagation, resource leaks, missing cleanup/finally blocks, inconsistent error strategies
Maintainability: Function/method length, cognitive complexity, deep nesting, code duplication, magic numbers/strings, dead code, unclear control flow
SOLID Principles: God classes, mixed responsibilities, rigid hierarchies, interface bloat, concrete dependencies, violation of contracts
DRY / KISS: Repeated logic, unnecessary abstractions, over-engineering, premature optimization, complex solutions for simple problems
Naming & Conventions: Unclear names, inconsistent naming style, misleading identifiers, language-idiomatic naming violations
Test Coverage: Missing tests for new/changed code, untested edge cases, brittle tests, missing integration tests, poor test isolation
Documentation: Missing/outdated API docs, undocumented public interfaces, missing changelog entries, misleading comments

DO NOT SKIP ANY CATEGORY

⚠️ STEP 5: Generate Report & Update Memory (REQUIRED)

YOU MUST ask user for preferred output format:

• Option A: Structured report → executive summary, categorized findings, action items → output to claudedocs/
• Option B: Inline comments → file:line feedback, PR-style
• Option C (Default): Both formats

DO NOT CHOOSE FORMAT WITHOUT USER INPUT

For EVERY issue in the output, YOU MUST provide:

1. Severity: Critical / Important / Minor
2. Category: Correctness / Error Handling / Maintainability / SOLID / DRY-KISS / Naming / Testing / Documentation
3. Description: What is wrong and why it matters
4. Fix: Concrete code example with improvement
5. File:line: Exact location (e.g., handler.go:142)

Format guidelines:

• Explain WHY (not just what)
• Show HOW to fix with examples
• Be specific with file:line references
• Be balanced (acknowledge good patterns)
• Educate, don't criticize
• Note language-specific idioms and conventions

DO NOT PROVIDE INCOMPLETE RECOMMENDATIONS

After completing the review, UPDATE PROJECT MEMORY:

Use memoryStore.update("code-reviewer", "{project-name}", ...) to create or update memory files:

1. project_overview: Languages, frameworks, architecture patterns, deployment info
2. common_patterns: Project-specific coding patterns and conventions discovered
3. known_issues: Recurring issues or anti-patterns found in this project
4. review_history: Summary of reviews performed with dates and key findings

Timestamps and staleness tracking are managed automatically by MemoryStore. See MemoryStore Interface for update() and append() method details.

---

Step 6: Generate Output

Create deliverables and save to /claudedocs/:

• Follow OUTPUT_CONVENTIONS.md naming: code-reviewer_{project}_{YYYY-MM-DD}.md
• Include all required sections
• Provide clear, actionable recommendations

Interface References

Interface	Usage in This Skill
ContextProvider	Load language/framework context dynamically based on detected stack
MemoryStore	Persist project review patterns, conventions, and history
SkillInvoker	Delegate to get-git-diff for change identification
ExecutionContext	Receive context from chained commands

Compliance Checklist

Before completing ANY review, verify:

• Step 1: Review scope identified using get-git-diff skill or user-specified files; languages and frameworks detected
• Step 2: Project memory loaded via memoryStore.getSkillMemory() and context detected via contextProvider
• Step 3: All relevant context files loaded via contextProvider.getAlwaysLoadFiles(), getConditionalContext(), and getCrossDomainContext() for each detected language
• Step 4: Deep review completed for ALL categories on changed code only
• Step 5: Output generated with all required fields AND project memory updated via memoryStore.update()

FAILURE TO COMPLETE ALL STEPS INVALIDATES THE REVIEW

Further Reading

• Quality Principles:
  • SOLID Principles: https://en.wikipedia.org/wiki/SOLID
  • Clean Code (Robert C. Martin)
  • Refactoring (Martin Fowler)
• Security:
  • OWASP Top 10: https://owasp.org/www-project-top-ten/
  • CWE Top 25: https://cwe.mitre.org/top25/
• Testing:
  • Test Pyramid: https://martinfowler.com/bliki/TestPyramid.html

Version History

• v1.0.0 (2026-02-12): Initial release — language-agnostic code review framework
  • Covers 8 review dimensions: correctness, error handling, maintainability, SOLID, DRY/KISS, naming, testing, documentation
  • Dynamic language/framework detection with contextProvider integration
  • Cross-skill discovery via memoryStore for related review insights