code-analyzer（代码分析器）

Code Analyzer

Analyze codebase to build comprehensive mental model for downstream operations.

Workflow Overview

1. Scan — Collect facts via bash script (deterministic)
2. Understand — Interpret structure and stack
3. Trace — Follow execution paths through abstraction layers (skip if greenfield)
4. Build — Construct dependency graph and mental model
5. Confirm — Output summary with key files list

Step 1: Scan Project

Run codebase scanner to collect facts:

.claude/skills/code-analyzer/scripts/scan-codebase.sh

Scanner auto-detects project root (git root or pwd) and collects:

• Structure: file count, extensions, configs, directories, src modules
• Markers: AICODE-NOTE, AICODE-TODO, AICODE-FIX with locations
• Git: branch, modified/added/deleted files

Outputs JSON. No external dependencies required.

Exclusions (automatic)

• node_modules,.git, dist, build
• pycache,.venv, venv
• ai-docs,.next,.nuxt, coverage,.cache

Step 2: Understand Structure

Interpret scan results to determine:

• Stack: Language(s) from extensions, framework from configs
• Entry points: Main/index/app files in directories
• Modules: Domain boundaries from src_modules or directories
• Conventions: Naming patterns, structure style

Step 3: Trace Patterns

Skip if greenfield project (no src files found in Step 1).

For existing codebases, trace how code actually works — not just what files exist:

3.1 Find similar features

• Grep for features with similar domain (e.g., if building "payments", find existing "orders" or "billing")
• Identify entry points: API routes, UI components, CLI commands

3.2 Trace execution paths

• Follow call chain from entry point through business logic to data layer
• Note data transformations at each hop with file:line references
• Document side effects and state changes encountered

3.3 Map abstraction layers

• Identify boundaries: presentation → business logic → data access
• Note which patterns are in use (repository, service layer, controller, middleware, etc.)
• Document cross-cutting concerns encountered (auth, logging, caching, error handling)

3.4 Identify reuse opportunities

• Shared utilities and helpers with 3+ consumers
• Existing patterns that new code should follow
• Modules that new feature should integrate with (not duplicate)

Step 4: Build Mental Model

Extract and internalize from scan results + tracing:

From structure:

• Stack: [language] | [framework] | [build-tool]
• Entry points with types
• Module list with inferred domains
• Directory organization

From tracing (if performed):

• Abstraction layers: [presentation] → [business] → [data]
• Design patterns in use with file:line examples
• Cross-cutting concerns and how they're implemented
• Reusable modules for new feature integration

From markers:

• AICODE-NOTE → Implementation context (why decisions were made)
• AICODE-TODO → Planned work (incomplete areas)
• AICODE-FIX → Known issues (from previous reviews)

From git:

• Current branch → feature context
• Changed files → review/focus scope

From reading key files:

• Import patterns → dependency relationships
• Shared modules → components with 3+ incoming connections
• Circular dependencies → architectural issues

Step 5: Confirm Readiness

Output minimal confirmation with key files list:

✅ Code context loaded: [project-name]
   Stack: [language] | [framework]
   Modules: [count] ([list])
   Patterns: [list of design patterns found, if traced]
   Markers: [N] NOTE, [N] TODO, [N] FIX

   Key files (read these for deep context):
   - [path] — [why this file matters]
   - [path] — [why this file matters]
   - ... (5-10 files max)

   Ready for: review | planning | documentation | agent-generation

Key files list — the 5-10 most important files for understanding the area being worked on. Calling agents/commands should read these files after Code Analyzer completes rather than re-scanning independently.

Error Handling

• Empty project: Report "No source files found"
• No git repo: Continue without git section (is_repo: false)
• Permission denied: Report file, continue with available
• No similar features found: Skip tracing, note "greenfield area — no existing patterns to follow"

Usage Notes

This skill prepares context for:

• Code review (scope, markers, dependencies)
• Implementation planning (patterns, reuse, architecture)
• Documentation generation (structure, stack)
• Agent creation (domains, boundaries)

Context remains in memory for entire conversation.