sowadalmughni

Agents Md Architect — Data skill for Claude Code

Data community

Detects which instruction file Claude Code actually loads (CLAUDE.md still wins if present) and generates a cross-tool AGENTS.md with real module boundaries and data flow rules, not just setup command.

How to install Agents Md Architect

This entry records only its repository, not the path inside it, so there is no exact command to give. Open sowadalmughni/agents-md-architect and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Agents Md Architect does

Alternatives in Data

LLM Redteam — LLM red-team corpus runner — fires categorized prompt-injection / jailbreak / system-prompt-leak / data-exfil 4.5k ★
Portaljs Architect — Recommend a data-portal architecture (storage, compute, catalog, access, hosting, metadata) from your needs, t 2.3k ★
Claude Code Memory Setup — Up to 71.5x fewer tokens per session on Claude Code with Obsidian + Graphify 947 ★

README

agents-md-architect

A Claude Code skill that generates and maintains a project's AI agent instruction file, and tells you the truth about a change most people have wrong.

**The correction:** Claude Code did not switch from CLAUDE.md to AGENTS.md. Version 2.1.277 (September 18, 2026) added AGENTS.md as a fallback: if no CLAUDE.md exists anywhere in the directory walk from your working directory upward, Claude Code reads AGENTS.md instead. If a CLAUDE.md exists, it still wins, unconditionally, with zero change in behavior.

**Why that distinction matters:** AGENTS.md is a real cross-tool standard, not a Claude Code feature. OpenAI released it in August 2025, contributed it to the Linux Foundation's Agentic AI Foundation, and by the time Claude Code caught up, over 60,000 open-source projects had already adopted it, read natively by Codex, Cursor, Devin, Gemini CLI, GitHub Copilot, Windsurf, Cline, Amp, and Aider. A repository with a CLAUDE.md is invisible to every one of those other tools. A repository with both files, maintained independently, drifts: one documented failure mode has Codex following one test command and Claude following a different one from a stale, disagreeing file.

This skill fixes the structural problem, not just the symptom. It detects exactly which file Claude Code is actually loading given the real walk rule (including the parts most people get wrong), generates a tool-agnostic AGENTS.md with real architectural content, and eliminates the drift by making AGENTS.md the single source of truth.

Install

# Via npx
npx skills add sowadalmughni/agents-md-architect

# Via Claude Code marketplace
/plugin marketplace add sowadalmughni/agents-md-architect
/plugin install agents-md-architect@sowadalmughni

# Manually — personal (follows you across projects)
mkdir -p ~/.claude/skills
git clone https://github.com/sowadalmughni/agents-md-architect.git ~/.claude/skills/agents-md-architect

# Manually — project-level
mkdir -p .claude/skill