vthink-agent-toolkit: Instructions file for Codex

AGENTS.md

This file provides guidance to AI coding agents (Claude Code, Cursor, Windsurf, Gemini CLI, OpenCode, etc.) when working with code in this repository.

Repository Overview

vthink-agent-toolkit is a collection of production-grade skills, agents, and workflows for AI coding agents. Skills and agents are plain natural-language procedural instructions — they work in any agent runtime that can read markdown files.

Structure

skills/<skill-name>/SKILL.md     # Skills — invoke by describing what you want
agents/<agent-name>/AGENT.md     # Agents — specialised sub-agents with a defined role
.claude/commands/                # Slash commands (Claude Code)
platforms/claude-code/           # Claude Code-specific: hooks, settings-templates, mcp, rules
playbooks/                       # Recipes for combining skills and agents
output-styles/                   # Response style files controlling tone and formatting

When Working in This Repo

If a task matches a skill, invoke it. Skills are located at skills/<skill-name>/SKILL.md.

Intent → Skill Mapping

What you're doing	Use this skill
Commit and push changes	commit-push or ship
Commit, push, and open a PR	commit-push-pr
Just need a commit message	give-me-a-commit-message
Check code against conventions	conventions-check
Write BDD / Gherkin scenarios	bdd-gherkin
Generate a QA test plan for a PR	qa-guide
Scaffold a CRUD endpoint	generate-endpoint
Anonymize a prompt for external tools	anonymize-prompt

Contribution Rules

• All skill/agent files must have name and description frontmatter only
• Skills go in skills/<name>/SKILL.md, agents in agents/<name>/AGENT.md
• Claude Code-specific assets (hooks, settings, MCP configs, rules) go in platforms/claude-code/
• Update CATALOG.md after every addition
• See CONTRIBUTING.md for full standards