project-setuplisted
Install: claude install-skill YoussefRashad/backend-agent-skills
# project-setup 3.1.0
One skill, one source, two tools. Produces a repository layout that Claude Code and Codex both follow:
```
AGENTS.md shared rules (canonical, <=150 lines, managed blocks)
CLAUDE.md @AGENTS.md + Claude-only notes
<dir>/AGENTS.md + CLAUDE.md optional local rules next to sensitive code (always as a pair)
.ai/manifest.json install state: version, ownership, hashes (written last, by apply.mjs)
.ai/agents/{reviewer,docs-sync}.md canonical agent definitions
.ai/context/*.md .ai/decisions/ navigation maps and evidenced decisions
docs/*.md conventions, database, integrations, observability
.claude/settings.json managed Read/command denies merged into user settings
.claude/agents/*.md Claude wrappers -> .ai/agents/*
.codex/agents/*.toml Codex wrappers -> .ai/agents/*
TODO.md everything uncertain, unverified, or out of scope
```
Read this file fully, then read references only when the step says so.
For Claude Code, you MUST start the session at the target repository root before setup.
Compare the original working directory in the supplied environment context with the target
root. If that context positively establishes a start below the root, you MUST ask for a
root restart and stop before file reads, Git commands or project tools.
As soon as a below-root start is positively established, including by discovery, you MUST
stop