---

name: mcp-structure-skill description: Guidance and snippets for creating/updating Codex skills tailored to the mcp-coding-agent project (AST/dependency tooling). Use when asked to build a new skill or refine one for this repo’s workflows and structure.

Purpose

Use this skill to design or refresh skills that help agents work on the mcp-coding-agent codebase. This project analyzes Python codebases by building AST structures with dependency tracking and optional LLM summaries.

Project Architecture

Core Components

1. ast_builder.py - Primary Stable Implementation

The main, production-ready AST builder with complete functionality:

• Data Classes: FunctionNode, ClassNode, FileNode with comprehensive fields
  • Tracks code, line numbers, summaries (code_summary, dependency_summary)
  • Inward dependencies (what code uses) and outward dependencies (what uses this code)
• DependencyVisitor: AST visitor extracting imports, function calls, class references, attributes
• LLM Integration: Optional OpenAI integration for code and dependency summarization
• Call Graph Building: _build_call_graph() maps function relationships across files
• Directory Walking: build_codebase_ast() processes all Python files in a directory tree
• CLI: python ast_builder.py <root_dir> [--summarize] [--json output.json]
• Requirements: Optional OPENAI_API_KEY for summaries

2. dependency_visitor.py - Standalone Visitor

Legacy/standalone version of DependencyVisitor class:

• Extracts imports, function calls, class references, attributes from AST nodes
• Used by second_builder.py
• Has identical structure to DependencyVisitor in ast_builder.py

3. second_builder.py - Alternative Builder (GitHub-focused)

Alternative implementation designed for string-based source code:

• AstBuilder Class: Works with source code strings (useful for GitHub content)
• Uses DependencyVisitor from dependency_visitor.py
• Similar data classes but slightly different field names
• Methods: build_file_ast(), extract_class(), extract_func()
• Less complete than ast_builder.py (no built-in call graph in class)

4. main.py - GitHub Repository Integration

Entry point for analyzing GitHub repositories:

• Integrates with devanalyzer.githubanalyzer to fetch code from GitHub
• Uses second_builder.py's AstBuilder class
• Implements _build_call_graph_from_dependencies() for GitHub-sourced content
• CLI: python main.py <repo_name> [--owner OWNER] [--branch BRANCH] [--no-summarize] [-o output.json]
• Requirements: GITHUB_TOKEN (environment variable), optional OPENAI_API_KEY
• Dependencies: Requires devanalyzer module with GitHub integration

5. agents.md - Project Documentation

High-level overview of repository structure, usage, and caveats

Quickstart for Skill Creation

1. Understand the ask: Determine if the skill is for:

   • AST/dependency analysis workflows
   • Local directory analysis (use ast_builder.py)
   • GitHub repository analysis (use main.py + second_builder.py)
   • LLM summarization features
   • Testing or validation

2. Load context appropriately:

   • Always check agents.md first for current project status
   • For local analysis: reference ast_builder.py
   • For GitHub integration: reference main.py and second_builder.py
   • For visitor patterns: reference dependency_visitor.py

3. Name the skill: hyphen-case, verb-led, <64 chars (e.g., analyze-ast-dependencies)

4. Structure the skill:

   • Frontmatter: name, description with clear triggers and scope
   • Content sections: Purpose, When to Use, Workflow, Examples
   • Keep under ~500 lines for maintainability

5. Decide resources:

   • scripts/: Only for truly reusable automation
   • references/: Brief domain notes, API references, examples
   • assets/: Rarely needed

6. Test scripts: If adding scripts, test locally with appropriate environment variables set

Common Usage Patterns

Local Directory Analysis

Copy# Basic AST analysis
python ast_builder.py /path/to/codebase

# With LLM summaries
python ast_builder.py /path/to/codebase --summarize

# Export to JSON
python ast_builder.py /path/to/codebase --summarize --json output.json

GitHub Repository Analysis

Copy# Requires GITHUB_TOKEN environment variable
export GITHUB_TOKEN='your_token_here'
export OPENAI_API_KEY='your_key_here'  # Optional

# Analyze repository
python main.py repo-name --owner username --branch main -o output.json

Key Data Structures

FileNode

• path, code_summary, dependency_summary, imports
• classes: List[ClassNode]
• functions: List[FunctionNode] (top-level only)

ClassNode

• name, lineno, end_lineno, code, code_summary, dependency_summary
• inward_dependencies: List[str]
• methods: List[FunctionNode]

FunctionNode

• name, lineno, end_lineno, code, code_summary, dependency_summary
• inward_dependencies: List[str] (what this uses)
• outward_dependencies: List[str] (what uses this)
• class_name: Optional[str] (if method)

Development Guidelines

• Stable foundation: Build on ast_builder.py for new local analysis features
• GitHub workflows: Extend main.py for repository-based features
• ASCII-only: Keep all code and docs ASCII-only
• Lightweight comments: Comment only non-obvious logic
• No destructive git: Avoid force pushes, resets, etc.
• Testing: No pytest suite yet; test manually or add targeted tests

SKILL.md Template (minimal)

Copy---
name: <skill-name>
description: <what it does + when to trigger, specific to mcp-coding-agent>
---

# Usage
- What problems this skill solves in this repo.
- When to invoke it (match triggers).
- Key steps (1-2-3) and any scripts/references to load.

# Resources
- scripts/<file> — what it does, how to run.
- references/<file> — when to read.

Packaging/Delivery

• Keep files ASCII-only. No extra READMEs/CHANGELOGs.
• If packaging is needed, zip the skill folder as <name>.skill mirroring structure (no special tooling required here).
• Remind users to restart Codex after installing the packaged skill.